PHYSICAL REVIEW D 75, 063512 (2007)

Scale dependence of halo and galaxy bias: Effects in real space

Robert E. Smith*
Department of Physics and Astronomy, University of Pennsylvania, 209 South 33rd Street, Philadelphia, Pennsylvania 19104, USA

Román Scoccimarro†
Center for Cosmology and Particle Physics, Department of Physics, New York University, New York, New York 10003, USA

Ravi K. Sheth‡
Department of Physics and Astronomy, University of Pennsylvania, 209 South 33rd Street, Philadelphia, Pennsylvania 19104, USA
(Received 22 September 2006; published 20 March 2007)
We examine the scale dependence of dark matter halo and galaxy clustering on very large scales
(0:01 < k
h Mpc1  < 0:15), due to nonlinear effects from dynamics and halo bias. We pursue a two line
offensive: high-resolution numerical simulations are used to establish some old and some new results, and
an analytic model is developed to understand their origins. Our simulations show: (i) that the z  0 dark
matter power spectrum is suppressed relative to linear theory by 5% on scales 0:05 < k
h Mpc1  <
0:075; (ii) that, indeed, halo bias is nonlinear over the scales we probe and that the scale dependence is a
strong function of halo mass. High mass haloes show no suppression of power on scales k <
0:07
h Mpc1 , and only show amplification on smaller scales, whereas low mass haloes show strong,
5%–10%, suppression over the range 0:05 < k
h Mpc1  < 0:15. These results were primarily estab-
lished through the use of the cross-power spectrum of dark matter and haloes, which circumvents the
thorny issue of shot-noise correction. The halo-halo power spectrum, however, is highly sensitive to the
shot-noise correction; we show that halo exclusion effects make this sub-Poissonian and a new correction
is presented. Our results have special relevance for studies of the baryon acoustic oscillation features in the
halo power spectra. Nonlinear mode-mode coupling: (i) damps these features on progressively larger
scales as halo mass increases; (ii) produces small shifts in the positions of the peaks and troughs which
depend on halo mass. We show that these effects on halo clustering are important over the redshift range
relevant to such studies 0 < z < 2, and so will need to be accounted for when extracting information
from precision measurements of galaxy clustering. Our analytic model is described in the language of the
‘‘halo model.’’ The halo-halo clustering term is propagated into the nonlinear regime using ‘‘1-loop’’
perturbation theory and a nonlinear halo bias model. Galaxies are then inserted into haloes through the
halo occupation distribution. We show that, with nonlinear bias parameters derived from simulations, this
model produces predictions that are qualitatively in agreement with our numerical results. We then use it
to show that the power spectra of red and blue galaxies depend differently on scale, thus underscoring the
fact that proper modeling of nonlinear bias parameters will be crucial to derive reliable cosmological
constraints. In addition to showing that the bias on very large scales is not simply linear, the model also
shows that the halo-halo and halo-dark matter spectra do not measure precisely the same thing. This
complicates interpretation of clustering in terms of the stochasticity of bias. However, because the shot-
noise correction is nontrivial, evidence for this in the simulations is marginal.

DOI: 10.1103/PhysRevD.75.063512 PACS numbers: 98.80.k

initial conditions, the energy content, shape, and evolution

I. INTRODUCTION
Statistical analysis of the large-scale structures observed
in galaxy surveys can provide a wealth of information
about the cosmological parameters, the underlying mass
distribution, and the initial conditions of the Universe. The
information is commonly extracted through measurement
of the two-point correlation function [1,2] or its Fourier
space analogue the power spectrum [3–7]. When further
combined with high precision measurements of the tem-
perature anisotropy spectrum from the cosmic microwave
background very strong constraints can be imposed on the
*Electronic address: res@astro.upenn.edu
†
Electronic address: rs123@nyu.edu
‡
Electronic address: shethrk@physics.upenn.edu
of the Universe [8].
For homogeneous and isotropic Gaussian random fields,
such as is supposed for the post inflationary density field of
cold dark matter (hereafter CDM) fluctuations, each
Fourier mode is independent, and thus all of the statistical
properties of the field are governed by the power spectrum.
However, nonlinear evolution of matter couples the Fourier
modes together, and power is transferred from large to
small scales [9,10]. Consequently, it is nontrivial to relate
the observed structures to the physics of the initial con-
ditions. Further, since one typically measures not the mass,
but the galaxy fluctuations, some understanding of the
mapping from one to the other is required. This mapping,
commonly referred to as galaxy bias, encodes the salient
physics of galaxy formation.

1550-7998= 2007=75(6)=063512(30) 063512-1 © 2007 The American Physical Society

SMITH, SCOCCIMARRO, AND SHETH
One last complication must be added: since galaxy
positions are inferred from recession velocities using
Hubble’s law, and because each galaxy possesses its own
peculiar velocity relative to the expansion velocity, a non-
trivial distortion is introduced to the clustering on all scales
from the velocity field. These velocity effects are com-
monly referred to as redshift space distortions. Thus one
must accurately account for nonlinear evolution of matter
fluctuations, bias, and redshift space distortions in order to
extract precise information from large-scale structure sur-
veys and this remains one of the grand challenges for
modern physical cosmology.
In this paper, we investigate the issue of bias in some
detail, through both numerical and analytic means. We
focus on real-space effects and reserve our results from
redshift space for a subsequent paper. Our numerical work
focuses on the generation of multiple realizations of the
same cosmological model, in two different box sizes. This
allows us to construct halo catalogues spanning a large
dynamic range in mass that are largely free from discrete-
ness fluctuations and the multiple realizations allow us to
derive errors that are ‘‘true errors from the ensemble.’’ We
use this data to show that not only is halo clustering on very
large scales scale dependent, but that the scale dependence
is a strong function of halo mass. These results are com-
pletely expected given the standard theoretical understand-
ing of dark matter haloes based on the ‘‘peak-background
split’’ argument [11–15].
Our analytic approach to modelling these trends can be
summarized as follows.
(i) Haloes are biased tracers of the mass distribution. To
describe this bias, we assume that the bias relation
between the halo density field and the dark matter
field is nonlinear, local, and deterministic. This al-
lows us to use the formalism of [13,16]. In order for
the local model to hold, one must integrate out small
scales where locality is almost certainly violated.
This can be done in real space by smoothing small-
scale fluctuations, or in Fourier space by considering
small wave numbers [17].
(ii) The underlying CDM density field is then propa-
gated into the nonlinear regime using standard
Eulerian perturbation theory techniques [10].
(iii) Galaxies are then assumed to form only in haloes
above a given mass [18] and these are inserted into
each halo using the ‘‘halo-model’’ approach [15,19–
22].
Because we write the perturbation theory (PT) evolved
halo density field as a series expansion we refer to this
method as ‘‘halo-PT’’ theory.
Our results are particularly relevant for studies which
intend to use the baryon acoustic oscillation feature (here-
after BAO) in the low redshift clustering of galaxies to
derive constraints on the dark energy equation of state [2].
The CDM transfer function that we have adopted through-
063512-2
PHYSICAL REVIEW D 75, 063512 (2007)
out contains a significant amount of BAOs, and we give an
accounting of the possible nonlinear corrections from mass
evolution and biasing that might influence the detection
and interpretation of such features. Previous work in this
direction has primarily focused on analysis of numerical
simulations [23–27], although several analytic works have
recently been presented: [28] derive the exact damping of
BAOs in the Zel’dovich approximation and calculate it in
the exact dynamics by resumming perturbation theory;
[29] consider real-space corrections to the power spectrum
from one-loop PT; [30] use the halo model, also in real
space, to explore systematics; [31,32] consider a model of
Lagrangian displacements fit to simulations.
In Sec. II we discuss how the approach we have devel-
oped here is complementary to and expands on these
studies. In Sec. III we discuss the numerical simulations
and present our measurements of scale dependence in the
dark matter, halo center, and halo-dark matter cross-power
spectra. We also present the evidence for large-scale non-
linear bias. Then in Sec. IV, we outline some key notions
concerning the halo model of large-scale structure, as this
is the frame work within which we work. In Sec. IV C we
describe the nonlinear bias model that we employ. In
Sec. IV D we use the 3rd order Eulerian perturbation theory
to describe the evolved Eulerian density field in terms of
the initial Lagrangian fluctuations. In Sec. V, we use the
3rd order halo density fields to produce an analytic model
for the 1-loop halo and halo-cross dark matter power
spectra. In Sec. VI we explore the predictions of the
analytic model for a range of different halo masses. In
Sec. VII we compare our analytic model to the nonlinear
bias seen in the numerical simulations. We use the analytic
model to examine the galaxy power spectrum in Sec. VIII,
and present our conclusions in Sec. X.
Throughout, we assume a flat Friedmann-Lemaı̂tre-
Robertson-Walker (FLRW) cosmological model with en-
ergy density at late times dominated by a cosmological
constant (
) and a sea of collisionless cold dark matter
particles as the dominant mass density. We take m 
0:27 and 
 0:73, where these are the ratios of the
energy density in matter and a cosmological constant to
the critical density, respectively. We use a linear theory
power spectrum generated from cmbfast [33], with
baryon content of b  0:046 and h  0:72. The normal-
ization of fluctuations is set through
8  0:9, which is the
initial value of the (root-mean-square) r.m.s. variance of
fluctuations in spheres of comoving radius 8h1 Mpc ex-
trapolated to z  0 using linear theory.
II. MOTIVATION
A number of recent papers have attempted to quantify
the scale dependence of galaxy bias. A subset of these have
forwarded simple analytic models to remove the scale-
dependent biases in the power spectrum estimator. We
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . .
discuss some of these below so as to set the stage for our
work.
A. Cole et al. (2005)
Based on the analysis of mock galaxy catalogues from
the Hubble volume simulation, these authors proposed a
simple analytic model to account for the nonlinear scale
dependence:
1  A2 k2
Pgg k  b2 PLin k : (1)
1  A1 k
The parameter A1  1:4, and A2 was allowed to vary over a
narrow range, which was then marginalized over in the
fitting procedure. When A1 k 1, this model has
Pgg k
b2 PLin k
1  A1 k  A2 k2 : (2) These authors used a set of three large cosmological
We show below that the bracketed terms are suggestive of
the P 
13  P22 terms from the dark matter perturbation
theory, but with incorrect dependence on k. In addition,
ignoring the fact that A1 may depend on galaxy type is a
serious inconsistency. For instance, our results indicate that
A1 for luminous red galaxy (LRG)-like galaxies is smaller
than 1.4.
A further concern regarding this model is that no ac-
counting for non-Poisson shot noise has been made. If
b2 PLin
1  A1 k2   A0 : (6)
galaxy formation takes place only in haloes, then galaxies
are not Poisson samples of the mass distribution. The
analytic model we develop shows that it is important to
account for this, and how. We are therefore skeptical about
the blind use of Eq. (1), particularly with regard to its use in
the analysis of LRGs [5,34].
B. Seo and Eisenstein (2005)
These authors examined the scale dependence of halo
bias in a large ensemble of low-resolution numerical simu-
lations. They proposed
Pgg k  b2 PLin k  A1 k  A2 k2  A0 ; (3)
which bares some similarity to that of Cole et al., as can be
seen by rewriting A1 ! A1 =b2 PLin  and A2 !
A2 =b2 PLin . The effective spectral index of the linear
power spectrum evolves from 0 > neff > 1 over the
range of k of interest, so we can think of PLin as being
approximately constant. Then, the inclusion of the b2 term
decreases the effective A1 as required, but it also decreases
the effective A2 ; our results indicate that this is
inappropriate.
There is an important difference between this model and
the previous one—the inclusion of the constant power
term A0 . This was introduced to account for ‘‘anomalous
power’’—by which was meant effects envisaged by
[35]—and/or non-Poisson shot noise following [36]. Our
analysis strongly supports the inclusion of this term.
063512-3
PHYSICAL REVIEW D 75, 063512 (2007)
C. Seljak (2001); Schulz and White (2006); Guzik,
Bernstein, and Smith (2006)
These authors explored the scale dependence of galaxy
bias in the halo model focusing their attention on the 1-halo
term. They showed that if this was taken simply as a non-
Poisson shot-noise correction, then it would be a signifi-
cant source of scale dependence in the large-scale galaxy
power spectrum. We have confirmed this in our study.
These studies lead one to suggest a base form of the kind:
Pgg k  b2 PLin k  A0 : (4)
On top of this base form we need to include modifications
that capture the true nonlinear evolution of the halo field.
D. Huff et al. (2006)
simulations to investigate the scale dependence of the
BAOs. They suggested
Pgg k  b2 PLin k exp
A1 k2   A0 ; (5)
where the exponential damping term was introduced to
account for halo profiles. Again examining the large-scale
behavior, A1 k 1, we see that this equation may be
rewritten
Pgg k
If 0 < A1 < 1 (the range they considered), then this for-
mula will suppress the power spectrum on scales k <
0:1h Mpc1 by a percent at most. If this model is to
account for the nonlinear corrections that we see, then
A1 > 1. However, the strong exponential damping makes
it unlikely that this model will properly characterize the
transition from the 2- to 1-halo term. This is because, as we
show below, the nonlinear evolution of halo centers in-
cludes an additional boost at intermediate k which this
model does not capture.
E. A necessary model
Our results suggest that a necessary model will have the
following properties: the model should be able to produce a
previrialization feature and a small-scale nonlinear boost
with k-dependencies motivated by physical arguments;
nonlinear corrections should depend on galaxy type; a
constant power term should be added to account for non-
Poisson shot noise. We therefore expect a reasonable start-
ing point for any empirical modelling of the large-scale,
scale dependence of the galaxy power spectrum to be
2
Pgg k; T  b2 T
PLin keA1 Tk  A2 TkmT 
1
 jWk=k j  A0 TjWk=k j  : (7)
n g
Our notation makes explicit that the coefficients have the
following properties: b > 0, A0 > 0, A1 > 0, A2 > 0, and
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
m  0, and that all depend on galaxy type T. The first term more. For the HR and LR simulations this corresponds to
is composed of two pieces: in the first piece, we have haloes with M > 4:0  1012 h1 M
and M > 4:0 
modeled the damping of BAOs using a Gaussian as derived 1013 h1 M
, respectively. We then constructed four non-
in [37,38] for the dark matter case; for the second piece, we overlapping subsamples of haloes with roughly equal num-
have added a k-dependent boost that models the power bers per subsample. The two low mass bins were harvested
added by mode-mode coupling and nonlinear bias. Our from the HR simulations and the two high mass ones were
simple power-law form (with two parameters A2 , m) is taken from the LR runs. Further details may be found in
meant to describe this effect over a restricted range of Table I.
scales. The fact that this term is additive as opposed to
multiplicative, is meant to emulate the fact that in PT this B. Mass, halo, and halo-mass power spectra
corresponds to the P 22 term which arises from the con-
volution of linear power on different scales and is therefore For each realization and each bin in halo mass we
measured the following quantities: the power spectrum of
smooth possessing no information on BAOs. For weakly
the dark matter P k; the power spectrum of dark matter
nonlinear scales it has a positive spectral index (note,
haloes Phh k; and the cross-power spectrum of dark matter
however, that in the limit k ! 0, m  4 is expected from
and dark matter haloes, Ph k. The power spectra may be
momentum conservation arguments). The second term
generally defined
corresponds to Poisson shot noise from unequal weighting
of haloes. The last term corresponds to the Poisson shot h k k0 i  23 D k  k0 P k; (9)
noise from the galaxy point distribution. We have included
filter terms Wk=k  and Wk=k  to indicate the damping where
due to density profiles, which will occur for k > 1=rvir . 
" #
i k ij P k Ph k
This function may be greatly simplified by examining the  k  ; P k  ;
h k Ph k Phh k
case k 1
h Mpc1 , for which it reduces to
(10)
Pgg k; T  b2 TfPLin
1  A1 Tk2   A2 TkmT g
with k and h k being the Fourier transforms of the
 A~0 T; k 1; (8) mass and halo density perturbation fields,
where the parameter A~0 T subsumes all sources of con- i x   i
1  i x; (11)
stant large-scale power.
where the index i again distinguishes between dark matter
III. SCALE-DEPENDENT HALO BIAS FROM and haloes, e.g.  1   and  2  h .
NUMERICAL SIMULATIONS We estimate the spectra through the conventional fast
fourier transform (FFT) method [42] (for a detailed dis-
A. Simulation details and halo catalogues cussion see [43]). The mean power and 1-
errors on the
We have performed a series of high-resolution, colli- spectra were estimated from the ensemble neglecting the
sionless dark matter N-body simulations, where N  5123 bin-to-bin covariances. Inspection of an estimate of the
equal mass particles. Each simulation was performed using covariance matrix from the 20 HR simulations showed
Gadget2 [39]; the internal parameter settings can be that this is reasonable. There is, however, a small degree
found in Table 1 of [40], where more details about the of off diagonal covariance, but the number of simulations
runs themselves are available. The initial conditions were was insufficient to make a precise estimate.
set up using the 2nd-order Lagrangian perturbation theory Figure 1 shows the three types of power spectra mea-
at redshift z  49 [40], with linear theory power spectrum sured from the ensemble of simulations for each of the four
taken from cmbfast [33], with the cosmological model
being the same as that used throughout this paper. We will TABLE I. Halo samples. Nreal is the number of independent
present results from two different box sizes: 20 smaller realizations. N h and n h are the ensemble average number and
higher resolutions box (hereafter HR) for which the vol- number densities of haloes in each mass bin.
ume is V  L3  512h1 Mpc3 , and 8 realizations of a
larger, lower resolution box (hereafter LR) for which L  Nreal L [Mpch1 ] N h [  104 ] n h [Mpc3 h3 ]
1024h1 Mpc box. LR Bin 1a 8 1024 3.6863 3:433 13  105
Haloes were identified in the z  0 outputs using the LR Bin 2b 8 1024 7.3530 6:8480  105
friends-of-friends algorithm with linking-length parameter HR Bin 3c 20 512 6.9287 5:1623  104
l  0:2. Halo masses were corrected for the error intro- HR Bin 4d 20 512 5.5415 4:1287  104
duced by discretization of the halo density structure [41]. a
Since the error in the estimate of the halo mass diverges as Mass bin 1  M > 1:0  1014 h1 M

b
Mass bin 2  1:0  1014 h1 M
> M > 4:0  1013 h1 M

the number of particles sampling the density field de- c
Mass bin 3  4:0  1013 h1 M
> M > 7:0  1012 h1 M

d
creases, we only study haloes containing 50 particles or Mass bin 4  7:0  1012 h1 M
> M > 4:0  1012 h1 M

063512-4
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 1 (color online). Halo power spectrum measurements and predictions, ratioed with a smooth ‘‘No-Baryon’’ dark matter power
spectrum, for four bins in halo mass: results for massive haloes (the top two panels) are from the LR simulations, whereas lower masses
(the two bottom panels) are from the HR simulations. Filled circles in the top, middle, and bottom sections of each panel show the
ensemble average nonlinear P k Phh k, and Ph , respectively. The open circles in the middle sections show Phh k with the
nonstandard shot-noise subtraction described in Appendix A. In all panels the linear theory dark matter, halo-halo, and halo cross-
power spectra are shown as dashed lines. The top panel also shows predictions from halofit (solid lines) [43] and 1-loop
perturbation theory (dot-dashed lines). Solid lines in the middle and bottom panels show our new analytic model, halo-PT.

063512-5
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
bins in halo mass described in Table I. The filled circles and TABLE II. Bias parameters for the halo samples. bST ST
1 , b2 , and
ST
associated error bars in the top, middle, and bottom sec- b3 are the first three nonlinear halo bias parameters derived
tions of each panel show P^  k, P^ hh kjM, and P^ h kjM. from the Sheth and Tormen model [14,15] averaged over halo
The open circles show the result of applying a nonstandard bins. b1 , b2 , and b3 are the parameters measured from the h -
scatter plots. bh is the large-scale bias parameter measured
shot-noise correction to P^ hh , which we describe in directly from Ph .
Appendix A.
To emphasize the nonlinear evolution of the spectra and bh bST
1 bST
2 bST
3 b1 b2 b3
the BAOs, in particular, we have divided each spectrum by Bin 1a 2:28  0:03 2.19 0.94 0:98 2.23 1.68 4:08
a smooth linear theory spectrum, which we shall refer to as Bin 2 1:49  0:04 1.53 0:30 0.94 1.41 0.04 1:29
our ‘‘No-Baryon’’ model. This was constructed by per- Bin 3 1:02  0:03 1.13 0:46 1.47 1.04 0:85 0.37
forming a chi-squared fit of the cmbfast transfer function Bin 4 0:87  0:03 0.98 0:44 1.44 0.91 0:74 0.55
data to the smooth transfer function model of [44],
a
See Table I for definition of bins.
Tk 
1  faq  bq1:5  cq2 gd 1=d : (12)
The derived parameters are: q  k=0:19, a  4:86, b  The dark matter power spectra in Fig. 1 show significant
4:81, c  1:72, d  1:18. deviations away from the linear theory prediction: at
To compare the halo spectra with the linear theory we 0:05 < k
h Mpc1  < 0:075, there is a suppression of
require estimates of the halo bias on very large scales, power relative to linear theory, whereas at k >
which we measure as follows. We begin by assuming 0:075
h Mpc1  there is an amplification. Perturbation
that the spectra can be written theory studies [10] refer to the suppression effect (caused
" #
1 bh by tidal terms) as previrialization. Recently, this has been
ij ij ij
P kjM  A PLin k; A  ; (13) understood in much more detail as a result of the damping
bh bhh 2
of linear features by nonlinearities, leading to an exponen-
where i and j denote the type of spectrum considered and tially decaying propagator that measures the loss of mem-
where, for reasons that will later be apparent, we distin- ory of the density field to the initial conditions [37,38].
guish between the bias from Ph and Phh . Hence, the Although the effect in the power is rather well known and
likelihood of obtaining an estimate of the power P^ ij in has been observed in recent numerical simulations of CDM
the lth k-bin is assumed to be an independent Gaussian spectra [4,43,46,47], our results constitute a rather precise
with dispersion
i measurement of this effect, with realistic errors drawn
 from the ensemble. A complete assessment of the damping

P^ ij 2
1 l  APLin kl  of linear features such as BAOs is done by studying the
LP^ ij
l ;
l jA; PLin   p
 exp  :
2
l 2
2l propagator [37,38] rather than the power spectrum. See
(14) [28] for further discussion on this.
The solid and dot-dashed lines show predictions based
Thus the combined likelihood of obtaining the data set on halofit [43] and on 1-loop PT. The PT results do
fP^ ij
l ;
l g can be written
well compared to the simulations on very large scales, but
for k > 0:07 they increasingly overpredict the power. We
Y
Ndat
L fP^ ij
l g; f
l gjA; PLin   LP^ ij
i ;
l jA; PLin : (15)
note also that PT appears to predict the previrialization
l1 feature in the simulations, adding additional support to the
On maximizing the likelihood function, we find the follow- claim that P^  k requires nonlinear corrections on the
ing estimator for the halo bias matrix scales of interest. However, qualitatively, it underpredicts
PNdat the magnitude of the effect. We note that halofit does
^ PLin kl P^ ij
l =
l
2
reasonably well at capturing the behavior for k <
A  Pl1 : (16)
Ndat
l1
PLin kl =
l 
2 0:07h Mpc1 , but it appears to underpredict the measured
data.
We construct error estimates through further differentiation Before moving on to Phh and Ph , we think it is worth
of the Gaussian likelihood function: noting that the BAOs in P on k > 0:1h Mpc1 have been
 2  NX
 erased. The large-box LR measurements show that the
@ logL 1=2 dat
PLin ki  2 1=2

A   : (17) third peak is gone, and the height of the second peak has
@A2 i1
i
dropped so that it appears more as a plateau. However, the
Lastly, since we observe scale-dependent nonlinear effects behavior at k < 0:05h Mpc1 appears unchanged.
in the matter power spectrum for k > 0:05h Mpc1 , we Discreteness corrections for P k have been studied in
only use modes with k < 0:04h Mpc1 in the fitting of the some depth [43]. However, for Phh , the appropriate correc-
amplitude matrix A. Estimates of the large-scale bias tion is more complicated because haloes are rare, highly
parameters bhh and bh are presented in Table II [45]. clustered, and spatially exclusive. In Appendix A we show

063512-6
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . .
that the standard Poisson shot-noise correction for the
cluster power spectrum results in negative power at high
k. This lead us to propose a new method for making the
shot-noise correction that accounts for exclusion, which we
discuss therein. The open circles in the middle sections of
each panel in Fig. 1 show the result of this new correction.
Filled circles show the uncorrected power, and stars show
the standard correction —clearly, the choice of correction
is crucial. Note that owing to the arbitrary normalization
things for the standard shot-noise method look better than
they actually are. Unfortunately, the residual uncertainties
in our new procedure prevent us from making strong state-
ments about the scale dependence of halo-halo clustering.
Whilst the discreteness correction is troublesome for Phh
it is almost negligible for Ph (the halo-model arguments
which follow allow us to quantify this). Our estimates of
Ph are shown in the bottom sections of the panels in
Fig. 1. Notice that the scale dependence of Ph , is a strong
function of halo mass. Ph for the most massive haloes
shows no deviations from linear theory until k >
0:7h Mpc1 . However, the previrialization feature appears
and gets enhanced as one goes to lower masses. Indeed, for
our lowest mass bin, Ph is sublinear until k >
0:15h Mpc1 .
This has important consequences for the BAOs. In the
highest mass bin (top-left panel), the oscillations in Ph at
k > 0:07h Mpc1 have been erased. However, the first
trough, at k  0:04h Mpc1 , is unaffected. For the next
mass bin (top-right panel), the first peak and trough are
unmodified, and the second peak is becoming noticeable.
This trend continues as we decrease mass; there is even a
hint of the third peak in the bottom panels. These mea-
surements indicate that nonlinear dynamics can erase os-
cillations on progressively larger scales as halo mass
increases and small displacements to the positions of the
peaks and troughs may occur; these will also be dependent
on halo mass. If the locations of these peaks and troughs
are to play an important role in constraining cosmological
parameters, our measurements suggest that understanding
and quantifying these displacements will be very
important.
Before continuing, we comment on the possible expla-
nation of these shifts through simple scatter from cosmic
variance. We remark that it is certainly possible to recon-
cile some of the shifts in the peak positions through this.
However, we draw attention to the fact that all of the points
PHYSICAL REVIEW D 75, 063512 (2007)
lations suggest changes of the order 1% to be acceptable;
and this increases to 3% for the HR simulations. If the
amplitudes for the theory curves are too low by 3%, then
some of these discrepancies may be alleviated. However, it
is unquestionable that nonlinear effects are present on these
scales and we must therefore firmly accept that it is likely
that these may cause some shifting of the harmonic series.
Only a wider and expanded numerical study will be able to
address and answer these questions more completely.
The solid lines in the middle and bottom sections of each
panel show predictions from the analytic model described
in the following sections. In all cases this model provides a
better description than does linear theory.
C. Scale dependence of the bias
Next, we examine the scale dependence of halo bias. We
will consider
v
u
u
t P^ ki  ; P^ h ki 
hh
b^ hh
NL ki   b^h
NL ki    ; (18)
P^ ki 

P^ ki 
as well as
s
P^ hh ki  P^ h ki 
b^ hh
Lin ki   ; b^h k
Lin i   : (19)
PLin ki  PLin ki 
For any particular realization the wave modes of the halo
and dark matter density fields are almost perfectly corre-
lated. Because the first set of estimators are derived from
taking the ratio of measured power spectra, they are in-
sensitive to this source of cosmic variance. In this sense,
the second set of estimators are nonoptimal. However, they
are the ones which will be used with real data, since P is
generally not observable.
In Fig. 2 we show the results of measuring these quan-
tities for the same halo-mass bins as in the previous sec-
tion. The top and bottom parts of each panel show (18) and
(19), respectively. The error bars, which were derived from
the ensemble to ensemble variations, are significantly
larger for (19) than for (18), as expected. The solid lines
show the predictions from the new analytic model de-
scribed in the next section.
For bhh we show both the shot-noise corrected (large
stars) and uncorrected (small stars) results. As was the case
for the halo-halo power spectra, we see that this correction
is important, so it must be known rather accurately. The
k  0:05 in the cross-power spectra of the low mass haloes estimators for bh are also shown (filled circles). Except for
are systematically lower than expected from the linear the highest mass bin, bhh  bh . Indeed, as we shall argue
theory. We also reiterate that the derived error bars are later, there are compelling theoretical reasons why the
the errors on the means for 20 realizations. One caveat is biases derived from Ph and Phh are not in fact the same,
that since the spectra are normalized by the very large-
scale modes of the power spectrum, where cosmic variance
errors are larger, we expect some small fluctuations in the
relative amplitudes of the theory predictions as more data
is acquired. Estimates of the error in the present LR simu-
and that, one generally expects bhh  bh . However owing
to the uncertainty regarding the shot-noise correction, no
firm statement can be drawn from the current data.
Two possible explanations why the highest mass halo
bin appears to behave differently are: First, if the shot-

063512-7
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 2 (color online). Large scale halo bias derived directly from N-body simulations for four bins in halo mass. The top sections of
each panel show the estimators b^ h ^hh
NL (solid points) and bNL with and without shot-noise correction (large and small stars, respectively).
The bottom panels show the same, but for b^hLin and b^ hh
Lin (See Eqs. (18) and (19) for definitions). The solid lines in each panel show the
predictions for the bias from our halo-PT model.

063512-8
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . .
hh
noise correction to P is too aggressive, we may have
underestimated the true Phh and therefore the bias.
Alternatively, we have made no shot-noise correction to
Ph ; if one should be applied (we think this is unlikely,
owing to the large number of dark matter particles), then
our current estimate of bh may be biased high.
The estimators based on Eq. (19) show more scale
dependence than those based on (18), especially for the
two high mass bins. As noted by [30], this is because the
BAOs are erased on larger and larger scales as higher and
higher mass haloes are considered. Thus on dividing the
halo spectra by a linear theory BAO spectrum, we are in
fact introducing scale dependence from the linear model.
For the highest mass haloes, b^h NL is constant at k <
0:04h Mpc1 , but it increases monotonically as k in-
creases. This is a direct consequence of the absence of
previrialization in Phh on intermediate scales and the rapid
onset of nonlinear power on smaller scales, compared to
P . The bias b^h
NL is flat for haloes with masses 4:0 
1013 < M
h1 M
 < 1:0  1014 , suggesting that small In the model the density fields of haloes and dark matter
clusters and group mass haloes are linearly biased tracers
of the nonlinear dark matter. The smallest two bins in halo
mass show the reverse trend: the bias decreases at large k,
by 10% compared to the approximately constant value at
smaller k.
IV. THEORETICAL MODEL
We now describe a model for interpreting the trends seen
in the previous section. Our discussion is based on the halo
model, which we briefly summarize below. See [48] for a
more detailed review.
PHYSICAL REVIEW D 75, 063512 (2007)
(i) All galaxies exist only in isolated dark matter haloes,
with more massive haloes hosting multiple galaxies.
In essence the model has been in existence for several
decades [12,18,49–53]. However it was not until the ad-
vent of large numerical N-body simulations and accurate
characterization of halo phenomenology that its true value
was realized. Namely, given an appropriate halo occupa-
tion distribution (HOD—i.e., a prescription for the number
and spatial distribution of galaxies within a halo) the model
successfully reproduces the real-space form of the two-
point correlation function of galaxies over a wide range of
scales. It predicts subtle deviations from a power-law
which have recently been seen in observations [54] and
provides a framework for describing the luminosity [55,56]
and environmental dependence of galaxy clustering
[57,58]. It also enables new tests of the CDM paradigm
to be constructed [59,60].
B. Power spectra
may be written as a sum over haloes,
X
Ni
i x   i
1  i x  Mj Uj x  xj jMj ; (20)
j
where i  f1; 2g distinguishes between haloes and dark
matter and N1  N and N2  Nh are the total number of
haloes and the number of haloes in some restricted range in
mass. Mj and xj are the mass and center of mass of the jth
halo and Uj  j xj =Mj is the mass normalized density
profile. Following [52], the power spectra P , Ph , and
Phh can be written as the sum of two terms:

A. The halo model of large-scale structure Pij k  Pij ij

1H k  P2H k: (21)
The halo model may be described by the simple state- The first term, Pij
1H , referred to as the 1-halo term, describes
ment: the intraclustering of dark matter particles within single
(i) All dark matter in the Universe is contained within a haloes; the second, Pij 2H , referred to as the 2-halo term,
distribution of CDM haloes, with masses drawn from describes the clustering of particles in distinct haloes. They
some mass function and with the density profile of have the explicit forms:
each halo being drawn from some universal stochas-
1 Z1
tic profile. Pij
1H k  i j dMnMM2 jUkjMj2  ij M; M;
The model attains its full potential when the second as-   0
sumption is stated: (22)

1 Z1Y 2
Pij
2H k  i j fdMl nMl Ml Ul kjMl gPhh
c kjM1 ; M2 ij M1 ; M2 ; (23)
  0 l1
where nM is the halo-mass function, which gives the number density of haloes with masses in the range M to M  dM,
per unit mass. The ij matrix carves out the halo density field to be considered, e.g. for haloes with mass M > Mcut the
matrix is


M1 M2  M1 M2  Mcut 
ij M1 ; M2   ;
M1  Mcut M2  M1  Mcut M2  Mcut 
where x is the Heaviside step function. More complicated halo selections can easily be described through the ij
notation. Lastly, Phh
c kjM1 ; M2  is the power spectrum of halo centers with masses M1 and M2 . This function contains all of

063512-9
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
the information for the interclustering of haloes; precise
knowledge of this term is required to make accurate pre-
dictions on large scales.
In principle, Phh
c kjM1 ; M2  is a complicated function of
M1 , M2 , and k. Initial formulations of the halo model
[15,20,21,61] assumed that it could be well approximated
by
Phh
c kjM1 ; M2   b1 M1 b1 M2 PLin k; (24)
where all the scale dependence is in PLin k, which is taken
from linear theory, and all the mass dependence is in the
scale-independent bias parameter b1 M [12,14]. In this
approximation, Phh c kjM1 ; M2  is a separable function of
M1 , M2 , and k. As we show below, comparison with
numerical simulations shows that this simple model over-
predicts power on very large scales and provides insuffi-
cient power on intermediate scales. In both cases, this is
about a 10% effect.
This discrepancy is not unexpected [22,62,63]. A simple
correction results from setting
Phh
c kjM1 ; M2   b1 M1 b1 M2 PNL k; (25)
where PNL is the nonlinear rather than the linear matter
power spectrum, and, in addition, imposing an exclusion
constraint:
hh
c rjM1 ; M2   1; r < rvir1  rvir2 ; (26)
where rvir is the virial radius of the halo.
The success of this approach is demonstrated by com-
paring P k measured in the z  0 output of the Hubble FIG. 3 (color online). Dark matter power spectrum measured
volume simulation [64] with the halo-model calculation. as a function of wave number measured from the z  0 time
The open and filled symbols in the top panel of Fig. 3 show slice of the Hubble volume simulation [64]. In the top panel the
the measurement before and after subtracting a Poisson points show the estimates of the dark matter power spectrum
shot-noise term (which is shown by the triple dot-dashed measured from the simulation, with and without a Poisson shot-
line.) The dot-short dash line shows the linear theory noise correction. The dot-dashed line shows the linear theory and
prediction, and the other two dot-dashed curves show the triple dot-dashed line shows the Poisson correction. The
two estimates of the 2-halo term: the one which drops dotted and dot-dashed curves show the 1- and 2-halo terms.
The thick dot-dashed curve shows the 2-halo term where PNL has
more sharply at large k is based on Eq. (25) and the other
been used instead of PLin . The bottom panel presents the ratio
one is based on the original approximation of Eq. (24). with respect to the halo model, but with Eq. (24) for the 2-halo
Since Eq. (25) requires the use of a nonlinear power term. The solid line shows the effect of the PNL modification.
spectrum, we used the one provided by [43].
The symbols in the bottom panel show the measure-
ments divided by the halo-model calculation which uses this, we note that the discrepancy between halofit and
Eq. (24) for the 2-halo term (i.e. the initial linear theory- the mass power spectra from our smaller box HR simula-
based approximation). Notice how they drop below unity at tions is somewhat puzzling. We highlight this issue for
k  0:1h Mpc1 . The solid line shows the halo-model further study, one possible explanation is the difference in
calculation which is based on Eq. (25)—it reproduces initial power spectra, on the other hand, also note that a
this previrialization feature well. calculation within the framework of renormalized pertur-
As an interesting aside, we note that the fitting formula bation theory [37] suggests that even for these simulation
halofit does very well at matching the previrialization volumes one expects small effects due to the absence of
feature. Whilst it is not apparent from the figure we also coupling to large scales.
point out that the transfer function of the Hubble volume To fix the small discrepancies which remain, some au-
simulation does contain BAOs; thus, our results demon- thors have advocated making the halo bias factors scale
strate that the fitting formula of [43] appears to be accurate, dependent [65], but the implementation has been based on
for this data, for BAO models to roughly 5%. In light of fitting formulae rather than fundamental theory. While

063512-10
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)
Eq. (25) appears to fare better than the original approxi- from the fact that hh xjRi  0, which leads to
mation (24), as we will soon show, in going from (24) to b2 2 b b
(25), one is making the assumption that halo bias is linear b0   h i  3 h3 i      n hn i: (32)
even when the mass density field is not. If this is not the 2 3! n!
case then the method is incorrect. In addition, there is an Thus, in general b0 is nonvanishing and depends on the
unpleasant circularity in requiring prior knowledge of hierarchy of moments. This allows us to rewrite Eq. (30) as
PNL k in order to predict PNL k. X
1
bi M
h xjR; M  f
xjRi  hi xjRig: (33)
i1 i!
C. Halo bias: The nonlinear local bias model
The discussion above makes clear that a rigorous treat- Nevertheless, we may remove b0 from further considera-
ment of the 2-halo term is currently lacking. This term tion by transforming to the Fourier domain, where it only
requires a description of how dark matter haloes cluster. contributes to k  0.
Whereas current models seek to describe halo clustering as The second constraint states that a sum over all halo
a biased version of dark matter clustering, the scale depen- density fields h xjM weighted by halo mass and abun-
dence of halo bias is still rather poorly understood [11– dance must recover the dark matter density field [15]. This
14,65–72]. In the following sections we develop a model to requires that

understand its main properties. In particular, we will dis- 1Z 1 i  1
cuss a general nonlinear, deterministic, local bias model for dMnMMbi M  : (34)
 0 i  0; 2; 3; . . .
dark matter haloes. This model is exactly analogous to that
derived for galaxy biasing by [16] and first applied to dark For CDM models whose initial density perturbations are
matter haloes by [13]. Gaussian random, the bias coefficients may either be de-
To begin, consider the density field of all haloes with rived directly through the ‘‘peak-background split’’ argu-
masses in the range M to M  dM, smoothed with some ment [11–14] or measured directly from N-body
filter of scale R. We now assume that this field can be simulations. Figure 4 shows the halo bias parameters up
related to the underlying dark matter field, smoothed with to third order, derived in the context of the Sheth-Tormen
the same filter, through some deterministic mapping and mass function; see [15] for the analytic expressions. We
that this mapping should apply independently of the pre- compare these with measurements from our simulations in
cise position, x, in the field: i.e. Appendix B.
A practical application of this method rests squarely
h xjR; M  F fM;Rg
xjR; (27) upon our ability to truncate the Taylor series at some
where the subscripts on the function F indicate that it
depends on the mass of the haloes considered and the
chosen filter scale. The filtered density field is
1 Z 3
xjR  d yyWjx  yj; R; (28)
V
Wjxj; R being some normalized filter. Taylor expanding
F fM;Rg about the point   0 yields
X bi MjR
F fM;Rg
xjR 
xjRi : (29)
i i!
We now assume that there is a certain filter scale above
which F fM;Rg is independent of both the scale considered
and also the exact shape of the filter function. Hence,
X
1
bi M
h xjR; M 
xjRi ; (30)
i0 i!
where the bias coefficients are

@i F fMg
 



bi M  
 : (31)
@ i  0
FIG. 4 (color online). First three halo bias parameters derived
The linear bias model has bi  0 for all i > 1. from the Sheth-Tormen [14] mass function as a function of halo
The bias coefficients from the Taylor series are not mass [15]. The solid line shows b1 , the dashed line b2 , and the
independent, but obey two constraints. The first arises dot-dashed line b3 .

063512-11
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
particular order. However, since the procedure that we have h
F1;2  b1 MWjkjRF1;2
adopted for doing this requires some further knowledge,
b2 M
we shall reserve our discussion until Sec. V B.  Wjq1 jRWjq2 jRF1 F2 ; (40)
2
D. Halo PT: Evolution of halo fields b2 M
h
F1;2;3  b1 MWjkjRF1;2;3 
Wjq1  q2 jR
We now evolve the halo density field as expressed by 3
Eq. (30) into the nonlinear regime via perturbation theory
 Wjq3 jRF1;2 F3  2cyc
techniques. For a short discussion of these methods see
Appendix C, and for a full and detailed review see [10]. b3 M
 Wjq1 jRWjq2 jRWjq3 jRF1 F2 F3 ;
The main idea that we require from perturbation theory is 6
that each Fourier mode of the density field may be ex- (41)
panded as a series, h
X where F1;...;j  Fjh q1 ; . . . ; qj jM; R. Thus Eqs. (37) and
k; a  Di1 ai k; (35) (38) can be used to describe the mildly nonlinear evolution
i of dark matter halo density fields to arbitrary order in the
where i k is the ith order Eulerian perturbation and dark matter perturbation, and Eqs. (39)–(41) make explicit
D1 a is the linear growth factor. Thus, on Fourier trans- the halo evolution up to 3rd order. Together, these ideas
forming the halo bias relation of Eq. (30), truncated at third define our meaning of the term halo-PT.
order, and on inserting the PT expansion from above, we It is now apparent that halo clustering studies which
arrive at (keeping up to cubic terms) assume a linear bias model and take the power spectrum
to be the fully nonlinear
P one, are effectively assuming that
h kjM; R  b1 M
1 kjR  2 kjR  3 kjR h kjM  b1 M i k, with bi M b1 M for i

b M Z d3 q1 1. However, for CDM, the peak-background split argument
 2
1 q1 jR1 k  q1 jR informs us that this never happens, unless the density field
2! 23
itself is linear [11–15]. We must therefore conclude that
 21 qjR2 k  q1 jR extrapolating the linear bias relation into the weakly non-
b M Z d3 q1 d3 q2 linear regime, without full consideration of the nonlinear-
 3 ity of halo bias is incorrect.
3! 26
 1 q1 jR1 q2 jR1 k  q1  q2 jR;
V. THE 1-LOOP HALO MODEL
(36)
A. Halo center power spectra
where i qjR  WjqjRi q. We next insert the solu- We now use the halo-PT to calculate the power spectrum
tions for each order of perturbation, which are presented in of halo centers in the mildly nonlinear regime. We define
Eq. (C4) of the Appendix, into Eq. (36). On rearranging the power spectrum of halo centers for haloes with masses
terms and collecting powers of , the mildly nonlinear M1 and M2 to be
density field of dark matter haloes may be written as a
PT series expansion of the dark matter density. This series hhc k1 jM1 hc k2 jM2 i  23 D k12 Phh
c k1 jM1 ; M2 :
is (42)
X
1 On inserting the halo-PT solutions for each order of the
h k; ajM; R  Dn1 a
h kjMn ; (37) perturbation we find that Phh
c k can be written as the sum
n1
of three terms
Z Qn fd3 qi 1 qi g Phh hh hh
c kjM1 ; M2   Pc;11 kjM1 ; M2   Pc;22 kjM1 ; M2 
h i1

 kjM; Rn 
23n3  2Phh
c;13 kjM1 ; M2 ; (43)
D

 kn Fnh q1 ; . . . ; qn jM; R; (38)
where
h
where
 kjMn is the nth order perturbation to the halo Phh h h
c;11 kjM1 ; M2   F1 kjM1 F1 kjM2 P11 k; (44)
density field, and where the short-hand notation

D kn  D k  q1      qn  has been used. The Z d3 q
functions Fnh q1 ; . . . ; qn jM; R are the halo-PT kernels, Phh
c;22 kjM1 ; M2   2 P11 qP11 jk  qj
symmetrized in all of their arguments. The first three 23
may be written in terms of the dark matter PT kernels:  F2h q; k  qjM1 F2h q; k  qjM2 ;
F1h  b1 MWjkjRF1 ; (39) (45)

063512-12
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)
Z d3 q and M2 nM2 , then the constraint equation (34) guarantees
Phh
c;13 kjM1 ; M2   6P11 k that they reduce to the standard 1-loop expression for the
23
PT power spectrum of dark matter [10]:
 P11 qF hh
13 k; qjM1 ; M2 : (46)
Here P11 k is equivalent to the linear theory power spec- P1-Loop k  P11 k  P22 k  P13 k: (48)
trum and
Strictly speaking the 1-loop power spectrum refers to
F hh 1 h h
13 k; qjM1 ; M2   2
F1 kjM1 F3 k; q1 ; q1 jM2  P22  P13 , we shall break convention and use Eq. (48) to
 F1h kjM2 F3h k; q1 ; q1 jM1 : define what we mean. Explicit details of the 1-loop ex-
pressions may be found in Appendix C 2.
(47)
The theory may be further developed by directly sub-
When these expressions are averaged over all halo masses, stituting the halo-PT kernels, given by Eqs. (39)–(41), into
weighted by the respective cosmic abundances M1 nM1  Eqs. (44)–(46). A little algebra shows that

Phh
c kjM1 ; M2 ; R
 b1 M1 b1 M2 P1-Loop k  b1;3 M1 ; M2 
2 RP11 k
jWkRj2
Z d3 q WqRWjk  qjR
 2b1;2 M1 ; M2  P11 qfP11 jk  qjF2 q; k  q  2P11 kF2 k; qg
23 WkR
b M b M  Z d3 q jWqRj2 jWjk  qjRj2
 2 1 2 2 P11 qP11 jk  qj; (49)
2 23 jWkRj2

where PkjR=jWkRj2  Pk: (51)

b i;j M1 ; M2   12
bi M1 bj M2   bj M1 bi M2 : (50)
Before continuing, we point out and answer an important
question that naturally arises at this junction: How does
one compare the filtered theory with the unfiltered obser-
vations? We forward the proposition that the unfiltered
nonlinear power spectrum can be recovered through the
following simple operation:
This is unquestionably true for an observed nonlinear field.
It will therefore also be true for the correct theoretical
model.
This rather lengthy expression may be more readily
digested through the examination of two limiting cases.
But first, notice the important fact that it is still a separable
product of mass dependent terms and scale-dependent
terms.
When the two halo masses are identical, then

Phh
c kjM1  M2 ; R
Z d3 q WqRWjk  qjR
2 2
 b 1 MP1 -Loop k  b 1 Mb 3 M
RP11 k  b 1 Mb 2 M P11 q
jWkRj2 23 WkR
 
 2P11 jk  qjF2 q; k  q  4P11 kF2 k; q

b22 M Z d3 q jWqRj2 jWjk  qjRj2

 P11 qP11 jk  qj: (52)
2 23 jWkRj2

This expression is equivalent to evolving the nonlinear, local, galaxy bias model [16] through Eulerian PT. This has been
explored by [73,74].
Second, consider the case where we integrate over one of the halo masses, say M2 , weighting by M2 and its abundance
nM2 . Equation (34) again insures that all terms involving b2 M2  and b3 M2  vanish, and so the resulting expression is
the 1-loop correction to the halo center-dark matter cross-power spectrum:

  Z d3 q WqRWjk  qjR
Ph
c kjM;R 1 2
 b1 MP1Loop k  b 3 M
RP11 k  b 2 M
jWkRj2 2 23 WkR
 P11 q
P11 jk  qjF2 q;k  q  2P11 kF2 k;q: (53)

063512-13
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
Inspection of these two limiting cases reveals three The third point noted above can be understood by con-
remarkable features: structing the linear bias, e.g. dividing Eqs. (55) and (56) by
(i) First, if the nonlinear bias parameters b2 M and P11 . On squaring the bias recovered from (56) and sub-
b3 M are nonvanishing then the bias on large scales tracting it from the bias from (55), we find
is not b1 M.
(ii) Second, the halo-halo spectrum has a term that cor- b22 M Z d3 q
f
bhh 2 h 2
Lin  
bLin  g 
P qjWqRj2 2
responds to constant power on very large scales, 2P11 k 23 11
whereas the cross spectrum does not.  2
34 1
Both of these points were independently noted by [73,74], 
4 R b2 M  b3 M ;
but for the case of nonlinear galaxy biasing (also see [35]). 21 2
(i) Third, the large-scale bias derived from the halo- (58)
dark matter cross-power spectrum is not b1 , nor is
it given by the bias derived from the halo-halo power We now see that, because Phh approaches a constant on
spectrum. very large scales, on dividing through by P11 kjM the bias
To see these points more clearly we take the k ! 0 limit function bhh diverges at the origin as 1=P1=2
11 diverges.
of Eqs. (49), (52), and (53): Figure 5 shows our expressions for bhh h
Lin and bLin (from

Eqs. (54) and (55)). In this particular case, we assume the
Phh kjM ; M ; R  b M b M P k 1 
2 R 34 nonlinear bias parameters derived from the Sheth-Tormen
c 1 2 1 1 1 2 11
21 mass function by [15]. We use a Gaussian filter, for which
  
b2 M1  b2 M2  1 b3 M1  WkRG   exp
kR2 =2. For RG  20h1 Mpc and for
   our fiducial cosmology, we find
2 RG  20
0:046. To
b1 M1  b1 M2  2 b1 M1 
 inspect the differences more closely, we take the ratio of
b M  b M b M 
 3 2  2 1 2 2 the predictions with respect to the tree-level theory, e.g.
q
b1 M2  2
Z dq 3 b1  Ph hh
11 =P11  P11 =P11 . The figure demonstrates two

P11 qjWqRj2 2 ; (54) of the points raised above. First the large-scale bias is not
23 simply b1 : halo-halo bias (solid through to dotted curves)

does not converge as one considers larger and larger scales;
68 b2 M
Phh
c kjM; R  b21 MP11 k 2
1 
R however bias from the cross-power spectrum is very close
21 b1 M to linear for all except the most massive haloes, where b2

b M b2 M Z d3 q
 3  2
b1 M 2 23

P11 qjWqRj2 2 ; (55)


34 b2 M
Ph
c kjM; R
2
 b1 MP11 k 1 
R
21 b1 M

1 b3 M
 : (56)
2 b1 M
These expressions make the first point noted above
trivially obvious: The large-scale bias is modulated by
the halo-PT correction terms, and these depend on the
nonlinear bias parameters b2 and b3 and also on the filtered
variance of fluctuations.
The second point noted above originates specifically
from the quadratic nonlinear bias terms found in
Eqs. (54) and (55), e.g. terms containing b2 M1 b2 M2 
and b22 M. For a linear power spectrum that obeys the limit
P11 k ! 0 ! 0, these expressions reduce to the constant
FIG. 5 (color online). 1-loop bias parameters bhh and bh in the
b22 M Z d3 q ultra large-scale limit. The solid through to dotted curves show
Phh
c k ! 0jM 
P11 qjWqRj2 2 :
2 23 bhh , measured on scales (kobs  f0:001; 0:005; 0:01;
(57) 0:05gh Mpc1 ), derived from the 1-loop halo-halo cross-power
spectrum as a function of halo mass. The triple dot-dashed curve
This term was discussed in great detail for the case of shows bh as derived from the 1-loop halo-dark matter cross-
galaxy biasing by [73]. power spectrum.

063512-14
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)
and b3 are very strongly rising functions (see Fig. 4). Hence for scales krvir 1, our Eqs. (22) and (23), at the
Second, it is now obvious that bhh and bh are not the 1-loop level in halo-PT, now take the forms:
same. Note also that the magnitude of the expected scale
Pij
1H kjR 
1 Z1
dependence: the bias varies by at most 5% when k is dMnMM2 ij M; M; (61)
]changed by an order of magnitude. The mass dependence jWkRj2   j 0
i

of bhh shown in Fig. 5 is simply driven by that of b22 M

(again see Fig. 4). As halo mass increases the bias slowly Pij
2H kjR 
1 Z1Y 2

increases until it reaches a maximum at M  1013 h1 M . fdMl nMl Ml g

It then decreases to M  1014 h1 M
after which it shoots
up dramatically for larger masses.
B. Convergence of the power spectrum
We now return to the issue of truncation and applicabil-
ity of the Taylor series expansion of the halo field. A first
requirement for the Taylor series to converge after a finite
number of terms is that the filter scale be large enough so
that the r.m.s. dark matter fluctuations be much less than
unity:
R 1. Considering the case where R is very
large and convergence occurs at first order, we then have:
h xjR  b1 M1 xjR. As the filter scale R is slowly
decreased the r.m.s. fluctuations in  increase and a larger
and larger number of terms are required to accurately map
the underlying bias function. Finally, as
R ! 1 all
terms in the series are required. At this point the method
has no merit.
Since a robust criterion for truncation is out of reach at
the present, we propose an ad hoc criterion for conver-
gence that must plausibly be obeyed, that is
b1 M
R < 1: (59)
In Appendix B we shall also discuss an empirical method
for testing convergence.
C. Returning to the halo model
We now translate these ideas back into the language of
the halo model. To begin, we shall restrict our attention to
the halo model in the large-scale limit, more precisely we
consider scales where UkjM  1. Since the Fourier trans-
form of the mass normalized profile may be written
Z rvir sin
kr
UkjM  d3 rUrjM ; (60)
0 kr
our large-scale condition simply becomes krvir 1. In the
above equation we have, for convenience, assumed spheri-
cal density profiles. If halo mass and virial radius are
related through M  4200r  3vir =3, then for the largest
collapsed objects in the Universe M
1015 h1 M
, and number. The different panels show results for different halo
the above condition translates to the inequality k masses and smoothing scales. In all four panels, we see
0:4h Mpc1 . If we assume an NFW density profile [75]
with concentration parameter c  6, then for k 
0:15h Mpc1 , we find that Uk
0:994. We therefore offset occurs for the cluster mass haloes, but here it may be
assume that this is an excellent approximation over the
scales that we are interested in.
jWkRj2  i  j 0 l1
Phh
c kjM1 ; M2 ; R
 ij M1 ; M2 ; (62)
jWkRj2
where we have explicitly included a filter on the 1-halo
term. (Recall that the halo center power spectrum in the 2-
halo term already includes such a filter.)
We now see that, because the bi M are the only mass
dependent functions, on insertion of Eq. (49) into Eq. (62)
the integrals over mass may be immediately computed.
Thus,
Pij
2H kjR  P
ij;hh
kjR; (63)
ij;hh
where P kjR is equivalent to Eq. (49), except that we
have replaced all of the mass dependent bias parameters by
the average ones: i.e.
R1
dMnMMbi MM  Mcut 
hbi i  0 R1 : (64)
0 dMnMMM  Mcut 
We note that if we wish to weight by halo number density
rather than mass density then we simply remove the mass
weighting in the numerator and denominator of hbi i.
VI. EVALUATION OF THE THEORY
In this section we present the results from the direct
computation of the 1-loop halo center expressions for Phh c
and Ph
c , as given by Eqs. (52) and (53), respectively. This
section is almost entirely pedagogical; we urge those who
are only interested in the direct comparison with the nu-
merical work to press on to Sec. VII.
Recall that it is necessary to adopt some filter scale R.
We have studied two choices: RG  20h1 Mpc and RG 
10h1 Mpc for which the linear theory, Gaussian filtered,
variances are
2 R  0:046 and 0.177, respectively. The
larger smoothing scale is required for the more massive
haloes.
A. Halo-dark matter cross-power spectra
In Fig. 6 we show the predictions for the scale depen-
dence of Ph at the 1-loop level, as a function of wave
that, as expected, there is a small (few percent) positive
offset from the linear theory bias value b1 . The largest
the case that, owing to the bias being large, the filter scale
that we have adopted for these objects may still be too

063512-15
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 6 (color online). The real-space cross-power spectrum of haloes and cold dark matter at the 1-loop level on large scales. The
four panels show predictions for haloes with masses: M  f1015 ; 1014 ; 1013 ; 1012 gh1 M . In the upper plot of each panel, the thick
curve represents the total spectrum as given by Eq. (53). The dashed curve gives the 1-loop contribution from the linear bias parameter
b1 term; the dotted curve and dot-dashed curves give the 1-loop contributions from the nonlinear bias parameters b2 and b3 , and the
thickness of the lines indicates their sign, with thick lines being positive and thin lines being negative contributions. The thin solid line
gives the smoothed, linearly biased, linear power spectrum. The lower plot of each panel shows the ratio of the total cross spectrum
with the smoothed linear theory cross spectrum.

063512-16
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 7 (color online). The real-space halo-halo power spectrum at the 1-loop level on large scales. The four panels show predictions
for haloes with masses: M  f1015 ; 1014 ; 1013 ; 1012 gh1 M . In the upper panels of each subfigure, the thick curve represents the total
spectrum as given by Eq. (52), the dashed curve represents the 1-loop contribution from the pure linear bias (b21 ) term, the dotted and
dot-dashed curves represent the 1-loop contributions from the b2 b1 and b3 b1 terms, respectively. The triple dot-dashed curve
represents the quadratic nonlinear bias term b22 . Lastly, the thin solid line denotes the smoothed, linearly biased linear theory spectrum.
As before, line thickness distinguishes positive and negative contributions. Bottom panels show the ratio with the linear spectrum.

063512-17
SMITH, SCOCCIMARRO, AND SHETH
small for adequate convergence. We also note that the
offset for the M  1014 h1 M
haloes is negligible. This spectra and the cross spectra. In particular: the large-scale
can be attributed to the fact that b2
b3
0 (see Fig. 4). bias is not given by b1 ; the highest mass halo spectrum
For the lower mass haloes the offsets are roughly 2% in
excess of linear.
Considering the predictions for the highest mass haloes,
we see that the spectrum is scale independent up to k 
0:07h Mpc1 , where the nonlinear amplification from the
P22 term becomes dominant. For this case, the absence of
the previrialization feature may be understood as follows:
First, we note that b1 and b2 are positive, whereas b3 is
negative. However, since b3 is an order of magnitude
smaller than the others, it plays no significant part in
determining the shape of the spectrum. The P13 term in
the 1-loop power spectrum, which is the main cause of the
previrialization feature, is thus overwhelmed by the action
of the quadratic bias b2 term. This suggests that Ph should
be scale independent up to k  0:07h Mpc1 .
Next, we collectively consider the predictions for the
, there is an upturn in
lower mass haloes, as they show many similar traits. First,
we find that when k < 0:01h Mpc1 , then the ratios of the
haloes this effect is not found, this owes to
1-loop to tree-level (or linear) spectra are flat. However, for
0:01h Mpc1 < k < 0:07h Mpc1 , significant scale de-
pendence is apparent: the previrialization feature is present
and it appears to become stronger as halo mass decreases.
On smaller scales still, the nonlinear boost from the P22
term amplifies the power spectrum and breaks all scale
independence. Interestingly, the onset of P22 is pushed to
smaller scales as halo mass decreases. These effects can be
understood as follows. For these objects b1 and b3 are
positive, whereas b2 is negative. On large scales, we see
that b2 and b3 are nearly equivalent, but b3 is slightly
dominant, and this results in a small positive correction.
Whereas on smaller scales this trend reverses and b2
becomes dominant. The overall correction is then negative
and this leads to the enhanced previrialization feature and
delay of the onset of P22 .
, and,
It is also interesting to note the imprint of the BAO
features in the ratios of the power spectra. The strength
of the signal appears to depend on halo mass and increases
as halo mass decreases. As we discussed in Sec. III C, this
can be attributed to the fact that nonlinear evolution sup-
presses BAOs on small scales, and on taking the ratio with
a linear theory spectrum, we are artificially introducing
oscillations.
B. Halo-halo power spectra
In Fig. 7 we show the predictions for the scale depen-
dence of Phh , at the 1-loop level, as a function of wave
number. The four panels show results for haloes with
masses in the same range as Fig. 6. Again, upper panels
show the contributions from each of the halo-PT terms in
Eq. (52), and where the sign of each contribution is dis-
tinguished through line thickness/color. Subpanels are as
before.
063512-18
PHYSICAL REVIEW D 75, 063512 (2007)
Some obvious similarities exist between the autohalo
shows no sign of the previrialization power decrement; the
nonlinear boost occurs increasingly at smaller scales as
halo mass decreases; with the exception of the highest
mass haloes, the ratio of the 1-loop spectra to the tree-level
spectra have BAOs imprinted. These effects may all be
understood through the explanations from the previous
subsection.
We also notice some important differences between Phh
and Ph . First, the addition of the quadratic nonlinear bias
term,
b2 M2 , modifies the results on the largest scales.
As was discussed in Sec. V, Phh 1-Loop becomes a white-noise
power spectrum as P11 ! 0, unless b2  0. In the figure,
the contributions from this term are denoted by the triple
dot-dashed lines. On considering all four panels and paying
special attention to the ratios, we see that, with the excep-
tion of the case M  1014 h1 M
power on scales k < 0:01h Mpc1 . For the M 
1014 h1 M
the fact that b2  0 (see Fig. 4).
Second, we note that on smaller scales the effect of the
b22 term is to boost the power across all scales. Thus the
previrialization power decrement is no longer a decrement
relative to the linear theory. However, relative to the power
measured on say k  0:01h Mpc1 , there is a very real
decrement. Moreover, the k-dependence of the terms mul-
tiplying b22 will make the decrement appear larger than we
would expect from the nonlinear matter spectrum.
C. Nonlinear evolution of BAOs
We now briefly consider how mode-mode coupling and
nonlinear biasing affect the evolution of the BAOs.
Figure 8 compares the 1-loop predictions for Phh with
the linear theory predictions. We again consider the set
of halo masses: M  f1015 ; 1014 ; 1013 ; 1012 gh1 M
to emphasize the evolution of the BAO features, we have
taken the ratio of each spectrum with the No-Baryon model
of Eq. (12). In addition, because we are purely interested in
the scale dependence, we renormalize all ratios to unity at
k  0:01h Mpc1 .
The top-left panel shows results for the highest mass
haloes: all but the first of the BAO troughs has been
removed through nonlinear evolution. This is a much more
aggressive nonlinear evolution than one expects from con-
siderations of the dark matter alone. Second, the trough
appears to have been displaced towards lower frequencies.
These effects arise because both the quadratic bias (b22 ) and
b2 b1 terms are positive, and so dominate over the negative
b3 and P13 terms. This fact, coupled with the injection of
power from P22 , means that the nonlinear boost occurs at
k  0:05h Mpc1 . Thus, the overall effect on the halo
spectrum is to shift the first trough to smaller k.
All the spectra of the lower mass haloes display a
previrialization feature, the strength of which increases as
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 8 (color online). Nonlinear evolution of the BAOs as traced by the halo-halo power spectrum for four different halo masses. The
solid line denotes the nonlinear (1-loop) spectrum as given by Eq. ; the dashed line denotes the linear theory (tree-level) power
spectrum. We have taken the ratio of each spectrum with a smooth No-Baryon model as described in the text, and have renormalized
each spectrum so that they are unity on a scale k  0:01h Mpc1 .

halo mass decreases. In addition, the number of peaks and
troughs which remain in the evolved spectra increases as
halo mass decreases, because the strong nonlinear ampli-
fication from the P22 term is delayed by the negative b2 b1
corrections. However, in contrast to the high mass haloes,
for the low mass haloes the first acoustic trough is shifted
towards larger k because, when 0:01 < k < 0:05, then the
negative correction from the b2 b1 term is dominant and so
subtracts power from the higher frequency side of the
oscillation. This acts to shift the overall pattern to higher
frequencies.
VII. COMPARISON WITH SIMULATIONS
In this section, we compare the predictions from the
theoretical model with the results from the numerical
simulations presented in Sec. III.
A. Halo center power spectra revisited
Returning to our analysis of Fig. 1, we are now in a
position to comment on the success of the halo-PT model
in comparison to the numerical simulations. The analytic
model was evaluated using the semiempirical bias parame-
ters that were determined as described in Appendix B.
Since here we are concerned purely with the scale depen-
dence of the spectra and not their overall amplitude, we
allowed the very large-scale normalizations to be consid-
ered as free parameters. These were fit for in exactly the
same way as was done for the linear theory models in
Eq. (13). Whilst this rescaling would not be necessary if
the bias parameters that we were adopting were precise and
accurate, since we have not been able to establish this in a
robust manner we feel that this approach is acceptable,
given that it has been equally applied to the linear theory.
Moreover, this is the method of analyzing real data.
These predictions are shown as the solid lines in Fig. 1.
Comparison with the measured Phh c , shows that the model
and the shot-noise corrected data show reasonable corre-
spondence over scales k < 0:07h1 Mpc. On smaller
scales the match is poor. However, the uncertainty in the
shot-noise correction makes it very difficult to draw firm
conclusions.
Considering Ph , we find that the analytic model is in
good agreement with the simulations over a much larger
range of scales. The model correctly captures the mass
dependence of the previrialization feature —low mass hal-
oes show greater loss of memory to the initial density
fluctuations. The shifting of the nonlinear boost to smaller
scales as halo mass decreases is also matched rather well.
To assess whether Phc;1-Loop is a better fit to the simula-
tion data than is linear theory, we have performed a like-
lihood ratio test, assuming that the likelihood functions are

063512-19
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
Gaussian. In this case, a necessary statistic for model tion using the halo model by changing the mass weighting
selection is for in the integrals to:
Lfxl ; yl gj
Pij
1-Loop max  nMM ! nMhNjMi;
LR  > 1; (65) (66)
Lfxl ; yl gj
Pij
Lin max  nMM2 ! nMhNN  1jMi;
where the subscript ‘‘max’’ refers to the parameter choices where hNjMi and hNN  1jMi are the first two factorial
in the models that maximize the likelihood. Restricting the moments of the halo occupation probability function
information to be k < 0:1h Mpc1 , we find, going from PNjM, which gives the probability for a halo of mass
low to high mass haloes, that LR  f1:06; 1:03; 1:05; 1:07g, M to host N galaxies. Use of the factorial moments of
respectively. PNjM subtracts off a term corresponding to the self-
correlation of galaxies, e.g. r ! 0  D r=n.  This
B. Halo bias revisited corresponds to the Poisson shot-noise term in Fourier
We now examine how well the halo-PT model does at space. Second, the mean density profile of dark matter is
matching the nonlinear scale-dependent bias of the halo changed to the mean density profile of galaxies in the halo:
centers as measured in the numerical simulations (Fig. 2). UkjM ! Ug kjM. Following the discussion in Sec. V C,
The top section of each panel shows that the analytic model it is a very good approximation to set Ug kjM  1 when
(solid lines correspond to bh hh
NL and bNL , respectively) cap- krvir 1. Third, the constant prefactors transform as
tures, qualitatively, the scale dependence of the bias. The 1= 2 ! 1=n 2g , where
model shows a bias that increases with k for the high mass Z
haloes but decreases with k for lower masses. However, n g  dMnMhNjMiM  Mcut : (67)
there are some notable discrepancies: The model under-
predicts and then overshoots the measured relationship for And finally, the galaxy bias parameters are
the most massive haloes; for the next bin in halo mass (top-
right panel), the measured bias is flat, whereas the model 1 Z
predicts a down turn after k  0:1h Mpc1 ; the model bgi  dMnMhNMibi M: (68)
n g
fares better in the two lowest mass bins, but the down
turn at high k is not seen in the data. However, we must
stress that, with the exception of the M  1014 h1 M These changes in Eqs. (61) and (62) yield the 1-loop halo-
haloes, the model outperforms linear theory, which would model prediction for the galaxy power spectrum.
predict constant bias on all scales. Explicitly:
Having extolled the virtues of our model we now draw Z
attention to its shortcomings. Whilst the predictions pro- Pgg
1H kjR  1
2 2
dMnMhNN  1jMiM  Mcut ;
vide a good match to the halo power spectra, they do not jWkRj n
simultaneously provide a good match to the scale depen- (69)
dence of the bias. If we reconsider our measurements of the
nonlinear matter power spectrum (upper sections in Fig. 1)
P , we see that the 1-loop model (dot-dashed lines) over- Pgg
2H kjR 1 Z Y
2
predicts the LR simulations on scales k > 0:05h Mpc1  fdMi nMi hNjMi iMi  Mcut g
jWkRj2 n 2 i1
and the HR simulations on scales k > 0:07h Mpc1 . It is
therefore unlikely that the model as presented here can be Phh
c;1-Loop kjM1 ; M2 ; R
made to work precisely.  ; (70)
jWkRj2

VIII. GALAXY POWER SPECTRUM

How does the scale dependence of the bias depend on
galaxy type? The answer has important consequences for
future galaxy surveys that will measure the clustering of
specific subclasses of objects. We may address this ques-
where again we have explicitly included a filter function on
the 1-halo term. On inserting our expression for Phhc from
Eq. (52) into the 2-halo term, and again noticing that in the
large-scale limit the mass integrals may be performed
directly, we find:

Pgg
2H g 2 g g 2
bg2 2 Z d3 q jWqRj2 jWjk  qjRj2
 f
b 1  P1-Loop k  b b
1 3
RP11 kg  P11 qP11 jk  qj
jWkRj2 2 23 jWkRj2
Z d3 q WqRWjk  qjR
 2bg1 bg2 P11 q fP11 jk  qjF2 q; k  q  2P11 kF2 k; qg; (71)
23 WkR

063512-20
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . .
The resulting expression for the galaxy power spectrum is
identical to that given by [73,74], with one important
difference —we include the large-scale constant power
originating from Pgg
1H kjR.
To study the expected differences between red and blue
galaxies, we use the parametric forms for hNMi mea-
sured by [76] in the semianalytic models of [77]:
   
M B M R
hNB jMi  0:7 ; hNR jMi  : (72)
MB MR
The blue galaxy parameters are: MB  4  1012 h1 M
; ratioed to the No-Baryon model. For the red galaxies, the
for haloes with masses in the range 1011  M=h1 M

4:0  1012 then B  0:0, for larger mass haloes B 
0:8. The red galaxy parameters are: MR  2:5 
1012 h1 M
; for haloes with masses greater than the cutoff
mass R  0:9. For the second moment of the HOD we
follow the model of Kravtsov et al. [78], so that hNN 
1jMi  hNjMi2  1. This makes PNjM sub-Poissonian
as suggested from the observations [20,56,63] and the
semianalytic models [15,19,76,79,80]; and second, this
choice allows the first moment alone to fully specify the
hierarchy of moments.
In practice, we use the models above but impose a lower
mass cutoff of Mmin  1012 h1 M
. This yields n B  also note that the BAOs are further suppressed and the
4:10  103 h3 Mpc3 , n R  7:93  103 h3 Mpc3 ,
B B B
fb1  1:20; b2  0:14; b3  1:06g, and fbR1 
R R B
1:39; b2  0:09; b3  0:89g. (This estimate of b2 agrees
with the observational determinations of similar galaxies
from the PSCz survey [81].) These values are easily under-
stood by noting how bi M depends on halo mass (e.g.
Fig. 4), the weightings given in Eq. (72), and recalling that
the halo-mass function declines exponentially at M >
1013 h1 M
. Notice that the red galaxy bias parameters halo term is not included, we see that the red and blue
are all positive, whereas bB2 for the blue galaxies is nega-
tive. Therefore, while we expect to see a previrialization
feature in PB k, we do not for the red galaxies.
The left panel of Fig. 9 shows the power spectrum of
blue and red galaxies evaluated using the 1-loop halo
model. The top section shows the individual contributions
from all the linear and nonlinear terms. Note that, for the
blue galaxies the nonlinear correction terms bB1 bB2 (dotted
lines) and bB1 bB3 (dot-dashed lines) are roughly the same
order of magnitude as the 1-halo term (thin line). However,
for the red galaxies, the 1-halo term (thin line) dominates
over the nonlinear bias corrections by factors of a few. For
both populations the quadratic bias terms
bg2 2 (triple dot-
dashed lines) appear to be negligible.
The middle section of the left-hand panel shows the
ratios of the 1-loop Pgg2H with the No-Baryon linear model
(from Eq. (12)). Again, since we are primarily interested in
the scale dependence of the bias, we renormalize all curves
063512-21
PHYSICAL REVIEW D 75, 063512 (2007)
to be unity at k  0:01h Mpc1 . As expected, the red
galaxy power spectrum appears to trace the linear theory
matter fluctuations (dashed line) very well when k <
0:07h Mpc1 . The nonlinear boost breaks this accordance
at larger k. However, for the blue galaxies, the scale
dependence is more complicated, having an increased
previrialization feature and a delayed nonlinear boost.
We also note that, for the red galaxies, the second and
third BAOs have been almost completely suppressed,
whereas only the third peak has been removed for the
blue galaxies. The peaks and troughs, however, appear to
be in the right places.
So far, we have neglected the contribution from the
constant power 1-halo term. In the bottom section of the
panel we now take this into account and show Pgg 1H  P2H
gg
agreement between the predictions and the linear theory
that was noticed before is now broken on much larger
scales, k  0:04h Mpc1 . The trough of the first BAO
has been shifted slightly to smaller k and the nonlinear
boost occurs at a larger scale. Because the 1-halo term is
about 5 times smaller for the blue galaxies, the modifica-
tions are not as severe. The addition of this term offsets the
suppression of power caused by the negative bB1 bB2 term
and the blue galaxies now appear to trace the linear theory
on scales k < 0:07h Mpc1 quite well. At larger k the
linear spectrum is a poor match to the predictions. We
second trough has been shifted to lower frequencies.
Because P1-Loop does not provide a very accurate model
for the true nonlinear power spectrum, we have studied the
effect of exchanging P1-Loop for the halofit [43] power
spectrum. The results are shown in the right-hand panel of
Fig. 9. Although the predictions are qualitatively very
similar, halofit predicts enhanced previrialization and
smaller nonlinear boosts. In the middle panel, where the 1-
galaxies do not match the linear theory as well on large
scales. In particular the blue galaxy power is suppressed on
all scales except the largest. We also see that the BAOs
have been better preserved. Although there are slight shifts
in the positions of the second trough and third peak.
However, the bottom panel shows that once the 1-halo
term has been included, the red galaxy predictions are
almost as before. The blue galaxies still show a reasonably
strong previrialization feature, but, because of the weak
nonlinear boost, they match the linear theory rather well
over nearly all the scales considered.
We note that this modification is not entirely self-
consistent and as such is not meant to be blindly trusted,
since the nonlinear bias terms are still derived from the 1-
loop halo-PT. However, we use this operation to highlight
that a more advanced understanding of the nonlinear power
spectrum does change the results quantitatively. This im-
plies that a more advanced model of the scale dependence
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 9 (color online). Scale dependence of the power spectrum of blue and red galaxies on very large scales. Top-left panel shows the
results obtained from the evaluation of Eqs. (69) and (71). The 1-halo term is represented by the thin solid lines. For the 2-halo term we
have the following contributions: b21 terms denoted by dashed lines; b2 b1 terms denoted by dotted lines; b3 b1 terms denoted by dot-
dashed lines; b22 terms denoted by triple dot-dashed lines. Note again for the 2-halo contributions, line thickness dictates sign. The
middle panel shows the ratio of the red and blue galaxy 2-halo terms with the No-Baryon linear model of Eq. (12). The dashed line
shows the linear theory model with BAOs. The bottom panel shows the same as the middle only this time we show the effect of
including the 1-halo term. The right-hand panel is similar to the left, only we exchange P1-Loop in the 2-halo term for the halofit
model [43].

of the bias will also further modify and improve the
predictions.
IX. SCALE-DEPENDENT BIAS AT z > 0
We have shown that the BAO harmonic series in the
power spectrum at (z  0) is affected by nonlinear ef-
fects from bias and gravitational evolution. Although
naively one might expect that at higher redshift non-
linear effects are less important, this is not necessarily
so. First, one must pick a criterion for how to compare
things at different redshifts. A natural choice is to use
objects of the same number density. In this case, as it
is well known, at higher redshifts objects of the same
number density are more biased, leading to stronger non-
linear bias effects, even though the dark matter has less
nonlinear evolution. Therefore, overall it is not clear a
priori whether the situation improves or not. Figure 10
shows the halo power spectra at z  1 and z  2 measured
from our 8 LR simulations, where the halo samples
were harvested so that they would have the same fixed
comoving number density as the Bin 1 sample at z  0
(see from Table I). The power spectra analysis was
identical to that as described in Sec. III. This figure
clearly demonstrates that the nonlinear bias effects that
are present at z  0 (Fig. 1), remain present in the high
redshift halo samples. In light of this, we anticipate that
low mass halo samples at higher redshift, constructed so
that M < M z > 0, will likewise show enhanced previri-
alization (M z is defined to be the halo mass at which

M  1). We reserve further details of this issue for
future work.

063512-22
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 10 (color online). Dark matter, halo-halo, and halo-dark matter cross-power spectra at z  1 (left panel) and z  2 (right panel)
as a function of wave number. Symbol styles are as in Fig. 1. At both epochs we computed the halo spectra using the same fixed
number density of haloes as those in the Bin 1 sample from z  0: n  3:43  105
h=Mpc3 . Nonlinear mass and bias corrections
remain present at the level of 5%–10% in the halo spectra.

high mass haloes this correction is sub-Poissonian,

X. CONCLUSIONS
In this paper we have explored in detail, through both
numerical and analytic means, the scale dependence of the
nonlinear dark matter, halo, and galaxy power spectra on
very large scales k < 0:15h Mpc1 . For our numerical
work we used an ensemble of 20 simulations in boxes of
side 512h1 Mpc and 8 simulations in boxes of side
1h1 Gpc. Each simulation contained more than 134  simulations.
106 particles. We have found that:
(i) The nonlinear matter power spectrum is suppressed
relative to the linear theory by 5%–10% on scales
between 0:05 < k=h Mpc1  < 0:075 at the 2-

level.
(ii) The halo-dark matter cross-power spectrum shows
clearly that the bias of halo centers is nonlinear on
very large scales. The form of the nonlinearity de-
pends strongly on halo mass: for high mass haloes no
previrialization suppression is seen; whereas there is
an apparent 10% suppression of power relative to
linear theory for lower mass haloes.
(iii) To make robust statements concerning their large-
scale clustering of haloes, it is essential to character-
ize the shot-noise correction to high precision. For
so the simple and widely used 1=n model must be
inappropriate. Halo exclusion effects lead to a plau-
sible explanation for this phenomenon, which we
used to motivate an alternate correction. However
further work is required to establish this robustly. In
addition the true answer may need to take into ac-
count the way in which haloes are identified in
(iv) The large-scale bias of Phh is not expected to be the
same as that of Ph due to nonlinear deterministic
bias; this complicates studies of stochastic bias. The
difficulty of performing the halo-halo shot-noise
correction means we are unable to make a strong
statement about the nonequivalence of bh and bhh .
(v) As wave number increases, low mass haloes are
increasingly antibiased and high mass haloes are
increasingly positively biased. Therefore, the non-
linear scale dependence of halo bias is not simply
due to the nonlinear evolution of the matter
fluctuations.
(vi) Baryon acoustic oscillation features in the power
spectrum are erased on progressively larger scales
as halo mass is increased. In addition, small shifts in

063512-23
SMITH, SCOCCIMARRO, AND SHETH
the positions of the higher-order peaks and troughs
occur which depend on halo mass.
In the second half of this paper we developed a ‘‘physi-
cal model’’ to explain and reproduce these results. The
model was constructed within the framework of the halo
model and we focused our attention on the clustering of the
halo centers. The halo-halo clustering term was carefully
propagated into the nonlinear regime using 1-loop pertur-
bation theory and a nonlinear halo bias model. Our model
can be summarized as follows: The density field of haloes
was assumed to be a function of the local dark matter
density field. Under the condition of small fluctuations, it
was then expanded as a Taylor series in the dark matter
density with nonlinear bias coefficients bi M [16]. The
density field was then evolved under 3rd order Eulerian
perturbation theory and this provided the 3rd order
Eulerian perturbed halo density field. We then used the
model to derive the halo-halo and halo-dark matter cross-
power spectra up to the 1-loop level. This lead to the
following conclusions:
(i) For bi nonvanishing, the effective bias on very,
very large scales for the halo spectra is not simply
b1 , but also depends on b2 , b3 and the variance
of fluctuations on scale R. The halo center power
spectrum contains a term that corresponds to con-
stant power on very large scales. This implies that
as k ! 0, halo bias should diverge as
PLin 1=2 .
The halo-dark matter cross-power spectrum does
not exhibit this behavior. The predicted bias from
this statistic approaches a constant value on large
scales.
(ii) When evaluated for a realistic cosmological model,
with nonlinear bias parameters taken from the Sheth-
Tormen mass function, the theory is in broad agree-
ment with numerical simulations for a wide range of
halo masses.
(iii) The nonlinear evolution of the BAOs was also exam-
ined. The model shows that nonlinear bias and non-
linear mode-mode coupling increasingly damp the
BAOs as halo mass is increased. In addition, the
positions of the peaks and troughs can be shifted
by small amounts which depend on halo mass.
Using the ensemble of simulations we constructed scatter
plots of the halo versus dark matter over densities, con-
tained in top-hat spheres of size R (see Appendix B). From
these, it was shown that for filter scales R < 60h1 Mpc
the bias was indeed nonlinear, and that, while the scatter
increased as R decreased, the mean of the relationship did
not change until R < 20h1 Mpc. We also examined
whether the nonlinear bias parameters derived from the
Sheth and Tormen mass function [14,15] provided a rea-
sonable match to the empirical halo bias.
Using semiempirical bias parameters as inputs for the
analytic model it was shown that the model well repro-
duced the scale dependence of the halo-dark matter cross-
063512-24
PHYSICAL REVIEW D 75, 063512 (2007)
power spectra in the simulations, and for all bins in halo
mass. However, it was only qualitatively able to reproduce
the scale dependence of the nonlinear halo bias.
The 1-loop halo center power spectrum was then in-
serted into the halo-model framework and defined the 1-
loop halo-model. This was used to predict the scale depen-
dence of blue and red galaxy power spectra. Plausible
models for the blue and red galaxy HODs were used and
the results showed complicated scale dependence.
Significant work still remains to be performed for this
analytic approach to be sharpened into a tool for precision
cosmology. Some possible improvements are: exchanging
the 1-loop matter power spectrum for an accurate analytic
fitting formula, i.e. after the fashion of halofit, but
designed purely for large scales; an application of the
new renormalized perturbation theory techniques
[37,38,82] coupled with the nonlinear bias model should
certainly produce better results.
The analytic and numerical results that we have devel-
oped here are concerned purely with the clustering in real
space. In a subsequent paper we shall extend our analysis
to explore the more realistic situation of the scale depen-
dence of dark matter, halo, and galaxy power spectra in
redshift space.
When this paper was nearing completion, a preprint
appeared [83] with similar calculations of how nonlinear
bias changes the power spectrum.
ACKNOWLEDGMENTS
R. E. S. would like to thank Peter Schneider for a very
thought provoking discussion that has led to the current
paper. We thank Jacek Guzik, Gary Bernstein, Bhuvnesh
Jain, and Laura Marian for many useful discussions during
this work. We would also like to thank Pat McDonald,
Daniel Eisenstein, Eiichiro Komatsu, and Martin White for
useful comments on the draft. R. E. S. thanks Joerg Colberg
and the Virgo Consortium for providing access to the
Hubble Volume and the VLS simulations. R. E. S. thanks
CCPP and NYU for their kind hospitality during part of
this research. R. S. would like to thank M. Manera for
useful discussions on halo exclusion. R. K. S. would like
to thank Martin White for many illuminating discussions
and the Aspen Center for Physics. We thank M. Crocce and
S. Pueblas for help regarding the numerical simulations
used here. R. E. S. and R. K. S. both acknowledge support
from the National Science Foundation under Grant
No. 0520647. R. S. is partially supported by NSF AST-
0607747 and NASA NNG06GH21G.
APPENDIX A: HALO DISCRETENESS
CORRECTIONS
1. Poisson model
It is necessary to correct the measured dark matter power
spectra for shot-noise errors. These arise through approx-
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)
imating the continuous CDM fluid by a point process. If the else they would have been linked as a bigger halo. As a
discretization of the density field obeys a Poisson process, result of this exclusion effect, the two-point correlation
i.e., dark matter particles are placed with probability / function of haloes drops dramatically from a value hh 
  xV, then the ‘‘discreteness’’ correction is [9]:

1 1 at r  re to hh  1 for r < re , where re is the exclu-
sion radius. This sharp drop of hh has an impact on the
Ptrue k  Pobs k  Pshot ; Pshot  1=n:
 (73) large-scale power spectrum, as we now show.
Since the number of particles in our simulations is large, Absent exclusion, the power spectrum would have been
5123 , this correction, on the scales of interest, is insignifi- Z 1 sinkr
cant, e.g. Pshot =P0:1h Mpc1   103 . However, one Phh k  hh 4r2 dr: (A1)
0 kr
must also correct Phh for discreteness. If the dark matter
haloes are also regarded as a Poisson sampling of the However, exclusion effects mean that
smoothed halo density field, then the correction will be Z 1 sinkr
the same but using the appropriate number density n h . In Phh k  hh 4r2 dr  vTH k; (A2)
Fig. 11 we show the effect of this standard correction on re kr
our halo power spectra and we plot the ensemble of the where vTH k  4=3r3e WTH kre  with WTH the Fourier
shot-noise corrected halo power. The negative power val- transform of a top-hat window in real space. Hence, the
ues that result at high k demonstrate that this model must difference in power due to exclusion is
not be exactly correct and therefore for accurate measure-
ments it must be modified in some way. We now discuss a Z re sinkr
possible explanation for this and propose a new shot Phh k  1  hh 4r2 dr: (A3)
0 kr
correction.
To estimate this, we must model hh at scales smaller than
the exclusion imposed by the friends-of-friends definition
2. Halo exclusion effects
of a halo. We do so simply by approximating hh by a
As we have seen Poisson sampling must not be exact for power-law hh  e r=re 
obtained from fitting to the
haloes, particularly those of large mass. A possible reason r  re measurements, where re is the exclusion scale. In
is that by using a friends-of-friends algorithm, one is this approximation
automatically imposing that haloes are never separated
by distances smaller than about the sum of their radii, or Phh k  vTH k  vke ; (A4)

FIG. 11 (color online). The effect of Poisson shot-noise correction on the halo-halo power spectra for the four bins in halo mass.

063512-25
SMITH, SCOCCIMARRO, AND SHETH
where, in the large-scale approximation kre 1,
4r3e 2r3e
vk ’  kr 2 : (A5)
3 
35 
 e correlated. The results of this exercise are tabulated in
Setting
 0 gives vTH in the same limit. Thus, we see Table II, which also shows the values of bi predicted
that exclusion makes the halo power spectrum smaller by
the amount given by Eq. (A4). In contrast, Poisson shot
noise makes the power larger by 1=n h . If Poisson noise is
subtracted from halo power spectra in simulations, then the
result becomes negative at high k, and so it may be that
adding back the power lost to exclusion will make the
power positive again. This is our procedure. Note that the
noise in Eq. (A4) is not white.
As a final note we reemphasize that all of these troubling
issues may, to a certain extent, be neatly side stepped, if we
measure the halo-dark matter cross-power spectrum. The
clear advantage of this approach is that the shot-noise is
dramatically reduced (there are many more particles than
haloes) and exclusion no longer plays a role.
APPENDIX B: MEASURING THE NONLINEAR
BIAS PARAMETERS
The accuracy of 1-loop halo PT depends on the accuracy
of the halo bias parameters b1 , b2 , and b3 . Figure 4 shows
these parameters derived from the conditional Sheth and
Tormen mass function [15]. There are a number of ways to
test the accuracy of the bias parameters. The most direct is
to smooth the halo and mass fields with a filter of scale R,
and to then plot h xjR versus xjR in the smoothed
fields (e.g. [68,84]). From the resulting scatter plot, one
then directly fits for the parameters bi up to some order,
subject to the constraint that hh i  hi  0.
Figure 12 shows such scatter plots for the four bins in
halo mass described earlier (Table I). In all cases, unlike
elsewhere in the paper, the fields were filtered with a ‘‘top-
hat’’ window function. Solid lines show results for four
filter scales RTH  f20; 30; 40; 50gh1 Mpc. The upper
and lower dotted lines show the 1-
errors about the
mean relation (solid lines). (Smoothing with a Gaussian
filter instead does not significantly change the mean, but
does slightly change the scatter around the mean relation.)
The figure shows that, above a certain scale, hh ji defines
a universal curve which is independent of R; it is this curve
which the bias coefficients are supposed to describe. The
figure also shows that the scatter around this mean curve
decreases as R increases. The assumption that bias is
deterministic is equivalent to assuming that this scatter,
this stochasticity, is negligible. Clearly, this is a reasonable
approximation only for large smoothing scales R, and R
must be bigger for the rarer, more massive haloes before
the bias can be called deterministic.
The main problem with this approach is that, when R is
large, then the distribution of  becomes sharply peaked
about its mean value (zero), so estimating the higher-order
bias coefficients becomes difficult —only the linear bias
063512-26
PHYSICAL REVIEW D 75, 063512 (2007)
term can be reliably measured. In effect, determination of
the bias coefficients corresponds to fitting a polynomial to
these mean curves, and the best-fitting coefficients will be
from the Sheth-Tormen mass function. Some of the dis-
crepancy is a consequence of the fact that the Sheth-
Tormen mass function underpredicts the abundance of
high mass haloes in the HR and LR simulations by up to
20% (see [40] for an explanation as to why this happens).
But note that, because of the correlation between the fitted
coefficients, the discrepancy with the theory values is
difficult to assess. The solid thick line in the figure shows
the mean bias relation derived from the Sheth-Tormen
mass function.
In light of these issues we use the following prescription
when generating the theory curves: b1 coefficients are to be
measured directly from the data as described in Sec. III B
and the parameters b2 and b3 are to be derived through
computing the following integrals over the relevant bin
width:
1 Z M2
b i M  dMnMbi M; (B1)
 1 ; M2  M 1
nM
where nM and bi M are the Sheth and Tormen mass
function and bias parameters. We shall refer to this ap-
proach as the semiempirical method. It shall be left to our
future work to provide a more self-consistent solution to
this problem.
APPENDIX C: EULERIAN PT
1. Real-space kernels
The growth of density inhomogeneities in an expanding
universe may be explored in the single stream approxima-
tion using the Eulerian fluid equations. In Fourier space the
equations governing the evolution of the fluctuations to the
density (k; t) and divergence of the peculiar velocity
field (k; t  r  v) are written [10]:
Z d3 q1 d3 q2
@k; t
 k; t  
@t 23

D k2 q1 ; q2 q1 ; tq2 ; t;
(C1)
@k; t 3
 Hak; t  m H 2 tk; t
@t 2
Z d3 q1 d3 q2

D k2 q1 ; q2 q1 ; tq2 ; t;
23
(C2)
where
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . . PHYSICAL REVIEW D 75, 063512 (2007)

FIG. 12 (color online). Scatter plot of halo number over density vs the over density of dark matter for four bins in halo mass. The
fluctuations were measured in top-hat spheres of radii RTH  f20; 30; 40; 50gh1 Mpc. The means and 1-sigma errors measured from
the ensemble are denoted by the solid, dashed, dot-dashed, and dotted lines, respectively. The solid thick line shows the predictions
from the Sheth and Tormen model. The thick triple dot-dashed lines show the results from the cubic fit in  as described in the text.

k  qi where Ha is the dimensionless Hubble parameter,

i;j  qi ; qj   ;
q2i D1 a  a is the linear growth factor, and fa 
(C3) d logD=d loga
5=9 m gives the velocity growth suppres-
k2 qi  qj 
i;j  qi ; qj   ; k  qi  qj : sion factor. The functions Fns and Gs
n are the PT kernels
2q2i q2j
for  and  symmetrized in all of their arguments, respec-
As shown in [10], for the case of the Einstein-de Sitter tively. The notation
D kn  D k  q1      qn 
model the density and velocity divergence fields may be was adopted. The first three symmetrized density kernels
expanded as a perturbation series and the solutions at each are:
order may then be written down in terms of the density and F1s q  1; (C6)
velocity perturbations from all lower orders: e.g.
X1 F2s q1 ; q2   14
5

1;2  2;1   271;2 ; (C7)
k  Dn1 an k;
n1
F3s q1 ; q2 ; q3   54
7 s

F1;2 s
3;12  F2;3 s
1;23  F3;1 2;31
Z Qn fd3 ki 1 ki g
n k  i1

D kn Fns k1 ; . . . ; kn ;  Gs s s
1;2 12;3  G2;3 23;1  G3;1 31;2 
23n3
(C4) 2
 27
Gs s s
1;2 12;3  G2;3 23;1  G3;1 31;2 :
X
1
(C8)
k  Hafa Dn1 an k;
n1
Z Qn fd3 ki 1 ki g The first three symmetrized velocity divergence PT kernels
n k  i1
3n3

D kn Gs
n k1 ; . . . ; kn ; are:
2
(C5) Gs
1 q  1; (C9)

063512-27
SMITH, SCOCCIMARRO, AND SHETH PHYSICAL REVIEW D 75, 063512 (2007)
Z dq Z
Gs 3 4
2 q1 ; q2   14
1;2  2;1   71;2 ; (C10) P13 k  6P11 k q 2
P q ^ 3 k; q; q
dqF
11
23

P11 kk3 Z 1
Gs 1 s s s
3 q1 ; q2 ; q3   18
F1;2 3;12  F2;3 1;23  F3;1 2;31  dx x P11 xk 42x2  100
2
25222 0
 Gs s s
1;2 12;3  G2;3 23;1  G3;1 31;2  158 12 3
 2  4  1  x2 3 7x2  2
 29
Gs s s
1;2 12;3  G2;3 23;1  G3;1 31;2 ;
x


x x

x1
(C11)  log ; (C14)
jx  1j
where we have adopted the compact notation
Z d3 q
P22 k  2 P11 qP11 jk  qj
F2 q; k  q2
23
Fis
1 ;...;in
 Fns qi1 ; . . . ; qin ; Z 1 dq Z1
(C12) 2 2
q2 P11 q dP11 k x; 
Gs s
i1 ;...;in  Gn qi1 ; . . . ; qin 
0 2 1


5 1 x x x; 
  
and 7 2 x;  x;  x

2 2
2 x
 ; (C15)
7 x; 
i1 ...in ;j1 ...jm  qi1      qin ; qj1      qjm ;
where x  q=k and where 2 x;   1  x2  2x. If
i1 ...in ;j  qi1      qin ; qj : (C13)
the P13 integral is truncated on large and small scales to
inhibit infrared and ultraviolet divergences, as may occur
Exact analytic solutions for arbitrary cosmological mod- for some power spectra, then identical constraints must
els have not yet been found. However, as was shown in also be placed on the P22 k. Explicitly, if we adopt
[85], under the assumption that Dn a /
D1 an and 
f
0:5 k < kfun
m , the solutions are identical to the Einstein- Pq  0 for ;
de Sitter solutions, but the growth factors are changed to k > kcut
those for the particular cosmological model in question. then the angular integral for P22 must necessarily have the
All other changes are small corrections [10]. new limits
Z1 Z 2
2. Evaluation of 1-loop expressions d ! ;
1 1
Some care is required in the numerical evaluation of the q
halo-PT expressions because fine cancellations can occur 2  min
1; k2  q2  k2cut =2kq; (C16)
between negative and positive terms. The approach that we q
adopt throughout this paper can be demonstrated through 1  max
1; k2  q2  k2fun =2kq:
the following example. Consider the 1-loop power spec-
trum for CDM given by Eq. (48). The PT corrections P13 Lastly, the particular values for the limits kfun and kcut were
and P22 may be analytically developed up to the following selected so that variance integrals,
2 R, would be con-
points: vergent within the finite range.

[1] E. Hawkins et al., Mon. Not. R. Astron. Soc. 346, 78
(2003).
[2] D. J. Eisenstein et al., Astrophys. J. 633, 560 (2005).
[3] M. Tegmark et al., Astrophys. J. 606, 702 (2004).
[4] S. Cole et al., Mon. Not. R. Astron. Soc. 362, 505 (2005).
[5] M. Tegmark et al., Phys. Rev. D 74, 123507 (2006).
[6] W. Percival et al., astro-ph/0608635.
[7] W. Percival et al., astro-ph/0608636.
[8] D. Spergel et al., astro-ph/0603449.
[9] P. J. E. Peebles, The Large-scale Structure of the Universe
(Princeton University, Princeton, NJ, 1980).
[10] F. Bernardeau, S. Colombi, E. Gaztañga, and R.
Scoccimarro, Phys. Rep. 367, 1 (2002).
[11] S. Cole and N. Kaiser, Mon. Not. R. Astron. Soc. 237,
1127 (1989).
[12] H.-J. Mo and S. D. M. White, Mon. Not. R. Astron. Soc.
282, 347 (1996).
[13] H.-J. Mo, Y.-P. Jing, and S. D. M. White, Mon. Not. R.
Astron. Soc. 284, 189 (1997).
[14] R. K. Sheth and G. Tormen, Mon. Not. R. Astron. Soc.

063512-28
SCALE DEPENDENCE OF HALO AND GALAXY BIAS: . . .
308, 119 (1999).
[15] R. Scoccimarro, R. K. Sheth, L. Hui, and B. Jain,
Astrophys. J. 546, 20 (2001).
[16] J. Fry and E. Gaztañaga, Astrophys. J. 425, 1 (1993).
[17] Several broad classes of the bias model may be defined:
local [86] and nonlocal [66,87–90]; linear or nonlinear
[16,19,20,73,91–93]; and deterministic or stochastic
[35,94]. With the exception of stochastic, local, linear
biasing, all these prescriptions result in some nontrivial
degree of scale dependence. Note that when weakly non-
linear scales are discussed, as we do in this paper, it may
not be entirely consistent to neglect nonlocality, which [43]
could potentially alter the predictions we present.
Nonlocal bias can be looked for in observations by using
higher-order statistics [81,95]. However, we may take
comfort in the fact that there is always a strong correlation [44]
between haloes and dark matter, since the former are built
from the latter. [45]
[18] S. D. M. White and M. Rees, Mon. Not. R. Astron. Soc.
183, 341 (1978).
[19] A. J. Benson, S. Cole, C. S. Frenk, C. Baugh, and C. Lacey, [46]
Mon. Not. R. Astron. Soc. 311, 793 (2000).
[20] J. A. Peacock and R. E. Smith, Mon. Not. R. Astron. Soc. [47]
318, 1144 (2000). [48]
[21] U. Seljak, Mon. Not. R. Astron. Soc. 318, 203 (2000). [49]
[22] A. Berlind and D. Weinberg, Astrophys. J. 575, 587 [50]
(2002). [51]
[23] A. Meiksin, M. White, and J. A. Peacock, Mon. Not. R. [52]
Astron. Soc. 304, 851 (1999).
[24] H.-J. Seo and D. Eisenstein, Astrophys. J. 598, 720 (2003). [53]
[25] H.-J. Seo and D. Eisenstein, Astrophys. J. 633, 575 (2005).
[26] M. White, Astropart. Phys. 24, 334 (2005). [54]
[27] E. Huff, A. E. Schulz, M. White, D. J. Schlegel, and M. [55]
Warren, Astropart. Phys. 26, 351 (2007).
[28] M. Crocce and R. Scoccimarro (unpublished). [56]
[29] D. Jeong and E. Komatsu, Astrophys. J. 651, 619 (2006). [57]
[30] J. Guzik, G. Bernstein, and R. E. Smith, astro-ph/0605594.
[31] D. J. Eisenstein, H.-J. Seo, E. Sirko, and D. Spergel, astro- [58]
ph/0604362.
[32] D. J. Eisenstein, H.-J. Seo, and M. White, astro-ph/ [59]
0604361.
[33] U. Seljak and M. Zaldarriaga, Astrophys. J. 469, 437 [60]
(1996).
[34] N. Padmanabhan et al., astro-ph/0605302. [61]
[35] R. Scherrer and D. Weinberg, Astrophys. J. 504, 607 [62]
(1998).
[36] U. Seljak, Mon. Not. R. Astron. Soc. 325, 1359 (2001). [63]
[37] M. Crocce and R. Scoccimarro, Phys. Rev. D 73, 063519
(2006). [64]
[38] M. Crocce and R. Scoccimarro, Phys. Rev. D 73, 063520
(2006).
[39] V. Springel, Mon. Not. R. Astron. Soc. 364, 1105 (2005).
[40] M. Crocce, S. Pueblas, and R. Scoccimarro, Mon. Not. R. [65]
Astron. Soc. 373, 369 (2006).
[41] M. S. Warren, K. Abazajian, D. E. Holz, and L. Teodoro, [66]
Astrophys. J. 646, 881 (2006).
[42] The dark matter particles/halo centers were assigned to a [67]
regular cubical grid using a fourth-order interpolation [68]
scheme, and each point on the grid was given equal
weight. The FFT of the gridded density field was then [69]
computed. Each resulting Fourier mode was corrected for [70]
063512-29
PHYSICAL REVIEW D 75, 063512 (2007)
convolution with the grid by dividing by the Fourier
transform of the mass assignment window function. The
power spectra on scale kl are then estimated by performing
the following sums,
1 XM
P^ ij kl   ji kl j kl  j2 ;
M l1
where M are the number of Fourier modes in a spherical
shell in k-space of thickness k, and  denotes complex
conjugation.
R. E. Smith, J. A. Peacock, A. Jenkins, S. D. M. White,
C. S. Frenk, F. R. Pearce, P. A. Thomas, G. Efstathiou, and
H. M. P. Couchman, Mon. Not. R. Astron. Soc. 341, 1311
(2003).
J. R. Bond and G. Efstathiou, Astrophys. J. 285, L45
(1984).
The procedure was also performed separately for the no-
baryon model spectrum; this was done in order for ratios
to be taken.
W. Percival et al., Mon. Not. R. Astron. Soc. 327, 1297
(2001).
V. Springel et al., Nature (London) 435, 629 (2005).
A. Cooray and R. K. Sheth, Phys. Rep. 372, 1 (2002).
J. Neyman and E. Scott, Astrophys. J. 116, 144 (1952).
P. J. E. Peebles, Astron. Astrophys. 32, 197 (1974).
J. McClelland and J. Silk, Astrophys. J. 217, 331 (1977).
R. Scherrer and E. Bertschinger, Astrophys. J. 381, 349
(1991).
R. K. Sheth and B. Jain, Mon. Not. R. Astron. Soc. 285,
231 (1997).
I. Zehavi et al., Astrophys. J. 608, 16 (2004).
X. Yang, H.-J. Mo, Y. P. Jing, F. C. van den Bosch, and Y.
Chu, Mon. Not. R. Astron. Soc. 350, 1153 (2004).
I. Zehavi et al., Astrophys. J. 630, 1 (2005).
R. Skibba, R. K. Sheth, A. Connolly, and R. Scranton,
Mon. Not. R. Astron. Soc. 369, 68 (2006).
U. Abbas and R. K. Sheth, Mon. Not. R. Astron. Soc. 372,
1749 (2006).
R. E. Smith and P. I. R. Watts, Mon. Not. R. Astron. Soc.
360, 203 (2005).
R. E. Smith, P. I. R. Watts, and R. K. Sheth, Mon. Not. R.
Astron. Soc. 365, 214 (2006).
C.-P. Ma and J. Fry, Astrophys. J. 543, 503 (2000).
R. K. Sheth, L. Hui, A. Diaferio, and R. Scoccimarro,
Mon. Not. R. Astron. Soc. 325, 1288 (2001).
X. Yang, H.-J. Mo, and F. van den Bosch, Mon. Not. R.
Astron. Soc. 339, 1057 (2003).
A. E. Evrard, T. J. MacFarland, H. M. P. Couchman, J. M.
Colberg, N. Yoshida, S. D. M. White, A. Jenkins, C. S.
Frenk, F. R. Pearce, J. A. Peacock, and P. A. Thomas,
Astrophys. J. 573, 7 (2002).
J. Tinker, D. Weinberg, Z. Zheng, and I. Zehavi,
Astrophys. J. 631, 41 (2005).
P. Catelan, S. Matarrese, and C. Porciani, Astrophys. J.
502, L1 (1998).
Y.-P. Jing, Astrophys. J. 503, L9 (1998).
R. K. Sheth and G. Lemson, Mon. Not. R. Astron. Soc.
304, 767 (1999).
Y.-P. Jing, Astrophys. J. 515, L45 (1999).
A. Kravtsov and A. Klypin, Astrophys. J. 520, 437 (1999).
SMITH, SCOCCIMARRO, AND SHETH
[71] T. Hamana, N. Yoshida, Y. Suto, and A. Evrard,
Astrophys. J. 561, L143 (2001).
[72] U. Seljak and M. Warren, Mon. Not. R. Astron. Soc. 355,
129 (2004).
[73] A. F. Heavens, S. Matarrese, and L. Verde, Mon. Not. R.
Astron. Soc. 301, 797 (1998).
[74] A. Taruya, Astrophys. J. 537, 37 (2000).
[75] J. Navarro, C. S. Frenk, and S. D. M. White, Astrophys. J.
490, 493 (1997).
[76] R. K. Sheth and A. Diaferio, Mon. Not. R. Astron. Soc.
322, 901 (2001).
[77] G. Kauffmann, J. M. Colberg, A. Diaferio, and S. D. M.
White, Mon. Not. R. Astron. Soc. 303, 188 (1999).
[78] A. Kravtsov, A. Berlind, R. Wechsler, A. Klypin, S.
Gottlober, B. Allgood, and J. Primack, Astrophys. J.
609, 35 (2004).
[79] A. Berlind, D. Weinberg, A. J. Benson, C. M. Baugh, S.
Cole, R. Davé, C. S. Frenk, A. Jenkins, N. Katz, and C. G.
Lacey, Astrophys. J. 593, 1 (2003).
[80] Z. Zheng, A. Berlind, D. Weinberg, A. J. Benson, C. M.
Baugh, S. Cole, R. Davé, C. S. Frenk, N. Katz, and C. G.
Lacey, Astrophys. J. 633, 791 (2005).
[81] H. Feldman, J. Frieman, J. N. Fry, and R. Scoccimarro,
063512-30
PHYSICAL REVIEW D 75, 063512 (2007)
Phys. Rev. Lett. 86, 1434 (2001).
[82] P. McDonald, astro-ph/0606028 [Phys. Rev. D (to be
published)].
[83] P. McDonald, Phys. Rev. D 74, 103512 (2006); 74,
129901(E) (2006).
[84] R. Casas-Miranda, H. J. Mo, R. K. Sheth, and G. Boerner,
Mon. Not. R. Astron. Soc. 333, 730 (2002).
[85] R. Scoccimarro, S. Colombi, J. N. Fry, J. A. Frieman, E.
Hivon, and A. Melott, Astrophys. J. 496, 586 (1998).
[86] P. Coles, Mon. Not. R. Astron. Soc. 262, 1065 (1993).
[87] N. Kaiser, Astrophys. J. 284, L9 (1984).
[88] A. Dekel and M. Rees, Nature (London) 326, 455 (1987).
[89] P. Catelan, C. Porciani, and M. Kamionkowski, Mon. Not.
R. Astron. Soc. 318, L39 (2000).
[90] T. Matsubara, Astrophys. J. 525, 543 (1999).
[91] R. Cen and J. Ostriker, Astrophys. J. 399, L113 (1992).
[92] R. Mann, J. A. Peacock, and A. F. Heavens, Mon. Not. R.
Astron. Soc. 293, 209 (1998).
[93] R. Cen and J. Ostriker, Astrophys. J. 538, 83 (2000).
[94] A. Dekel and O. Lahav, Astrophys. J. 520, 24 (1999).
[95] J. A. Frieman and E. Gaztañaga, Astrophys. J. 425, 392
(1994).